[BACK]Return to uvm_map.c CVS log [TXT][DIR] Up to [cvs.NetBSD.org] / src / sys / uvm

Annotation of src/sys/uvm/uvm_map.c, Revision 1.204.2.9

1.204.2.9! yamt        1: /*     $NetBSD: uvm_map.c,v 1.204.2.8 2008/02/27 08:37:09 yamt Exp $   */
1.1       mrg         2:
1.98      chs         3: /*
1.1       mrg         4:  * Copyright (c) 1997 Charles D. Cranor and Washington University.
1.98      chs         5:  * Copyright (c) 1991, 1993, The Regents of the University of California.
1.1       mrg         6:  *
                      7:  * All rights reserved.
                      8:  *
                      9:  * This code is derived from software contributed to Berkeley by
                     10:  * The Mach Operating System project at Carnegie-Mellon University.
                     11:  *
                     12:  * Redistribution and use in source and binary forms, with or without
                     13:  * modification, are permitted provided that the following conditions
                     14:  * are met:
                     15:  * 1. Redistributions of source code must retain the above copyright
                     16:  *    notice, this list of conditions and the following disclaimer.
                     17:  * 2. Redistributions in binary form must reproduce the above copyright
                     18:  *    notice, this list of conditions and the following disclaimer in the
                     19:  *    documentation and/or other materials provided with the distribution.
                     20:  * 3. All advertising materials mentioning features or use of this software
                     21:  *    must display the following acknowledgement:
                     22:  *     This product includes software developed by Charles D. Cranor,
1.98      chs        23:  *      Washington University, the University of California, Berkeley and
1.1       mrg        24:  *      its contributors.
                     25:  * 4. Neither the name of the University nor the names of its contributors
                     26:  *    may be used to endorse or promote products derived from this software
                     27:  *    without specific prior written permission.
                     28:  *
                     29:  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
                     30:  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
                     31:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
                     32:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
                     33:  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
                     34:  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
                     35:  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
                     36:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
                     37:  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
                     38:  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
                     39:  * SUCH DAMAGE.
                     40:  *
                     41:  *     @(#)vm_map.c    8.3 (Berkeley) 1/12/94
1.3       mrg        42:  * from: Id: uvm_map.c,v 1.1.2.27 1998/02/07 01:16:54 chs Exp
1.1       mrg        43:  *
                     44:  *
                     45:  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
                     46:  * All rights reserved.
1.98      chs        47:  *
1.1       mrg        48:  * Permission to use, copy, modify and distribute this software and
                     49:  * its documentation is hereby granted, provided that both the copyright
                     50:  * notice and this permission notice appear in all copies of the
                     51:  * software, derivative works or modified versions, and any portions
                     52:  * thereof, and that both notices appear in supporting documentation.
1.98      chs        53:  *
                     54:  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
                     55:  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
1.1       mrg        56:  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
1.98      chs        57:  *
1.1       mrg        58:  * Carnegie Mellon requests users of this software to return to
                     59:  *
                     60:  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
                     61:  *  School of Computer Science
                     62:  *  Carnegie Mellon University
                     63:  *  Pittsburgh PA 15213-3890
                     64:  *
                     65:  * any improvements or extensions that they make and grant Carnegie the
                     66:  * rights to redistribute these changes.
                     67:  */
                     68:
1.114     lukem      69: /*
                     70:  * uvm_map.c: uvm map operations
                     71:  */
                     72:
                     73: #include <sys/cdefs.h>
1.204.2.9! yamt       74: __KERNEL_RCSID(0, "$NetBSD: uvm_map.c,v 1.204.2.8 2008/02/27 08:37:09 yamt Exp $");
1.114     lukem      75:
1.21      jonathan   76: #include "opt_ddb.h"
1.6       mrg        77: #include "opt_uvmhist.h"
1.169     petrov     78: #include "opt_uvm.h"
1.31      tron       79: #include "opt_sysv.h"
1.1       mrg        80:
                     81: #include <sys/param.h>
                     82: #include <sys/systm.h>
                     83: #include <sys/mman.h>
                     84: #include <sys/proc.h>
                     85: #include <sys/malloc.h>
1.25      thorpej    86: #include <sys/pool.h>
1.104     chs        87: #include <sys/kernel.h>
1.112     thorpej    88: #include <sys/mount.h>
1.109     thorpej    89: #include <sys/vnode.h>
1.204.2.6  yamt       90: #include <sys/lockdebug.h>
1.204.2.7  yamt       91: #include <sys/atomic.h>
1.1       mrg        92:
                     93: #ifdef SYSVSHM
                     94: #include <sys/shm.h>
                     95: #endif
                     96:
                     97: #include <uvm/uvm.h>
1.151     chs        98: #undef RB_AUGMENT
                     99: #define        RB_AUGMENT(x)   uvm_rb_augment(x)
1.21      jonathan  100:
                    101: #ifdef DDB
                    102: #include <uvm/uvm_ddb.h>
                    103: #endif
                    104:
1.204.2.1  yamt      105: #if defined(UVMMAP_NOCOUNTERS)
1.169     petrov    106:
1.204.2.1  yamt      107: #define        UVMMAP_EVCNT_DEFINE(name)       /* nothing */
                    108: #define UVMMAP_EVCNT_INCR(ev)          /* nothing */
                    109: #define UVMMAP_EVCNT_DECR(ev)          /* nothing */
                    110:
                    111: #else /* defined(UVMMAP_NOCOUNTERS) */
                    112:
1.204.2.2  yamt      113: #include <sys/evcnt.h>
1.204.2.1  yamt      114: #define        UVMMAP_EVCNT_DEFINE(name) \
                    115: struct evcnt uvmmap_evcnt_##name = EVCNT_INITIALIZER(EVCNT_TYPE_MISC, NULL, \
                    116:     "uvmmap", #name); \
                    117: EVCNT_ATTACH_STATIC(uvmmap_evcnt_##name);
                    118: #define        UVMMAP_EVCNT_INCR(ev)           uvmmap_evcnt_##ev.ev_count++
                    119: #define        UVMMAP_EVCNT_DECR(ev)           uvmmap_evcnt_##ev.ev_count--
                    120:
                    121: #endif /* defined(UVMMAP_NOCOUNTERS) */
                    122:
                    123: UVMMAP_EVCNT_DEFINE(ubackmerge)
                    124: UVMMAP_EVCNT_DEFINE(uforwmerge)
                    125: UVMMAP_EVCNT_DEFINE(ubimerge)
                    126: UVMMAP_EVCNT_DEFINE(unomerge)
                    127: UVMMAP_EVCNT_DEFINE(kbackmerge)
                    128: UVMMAP_EVCNT_DEFINE(kforwmerge)
                    129: UVMMAP_EVCNT_DEFINE(kbimerge)
                    130: UVMMAP_EVCNT_DEFINE(knomerge)
                    131: UVMMAP_EVCNT_DEFINE(map_call)
                    132: UVMMAP_EVCNT_DEFINE(mlk_call)
                    133: UVMMAP_EVCNT_DEFINE(mlk_hint)
                    134:
                    135: UVMMAP_EVCNT_DEFINE(uke_alloc)
                    136: UVMMAP_EVCNT_DEFINE(uke_free)
                    137: UVMMAP_EVCNT_DEFINE(ukh_alloc)
                    138: UVMMAP_EVCNT_DEFINE(ukh_free)
1.169     petrov    139:
1.87      enami     140: const char vmmapbsy[] = "vmmapbsy";
1.1       mrg       141:
                    142: /*
1.204.2.7  yamt      143:  * cache for vmspace structures.
1.25      thorpej   144:  */
                    145:
1.204.2.7  yamt      146: static struct pool_cache uvm_vmspace_cache;
1.25      thorpej   147:
1.26      thorpej   148: /*
1.204.2.7  yamt      149:  * cache for dynamically-allocated map entries.
1.26      thorpej   150:  */
                    151:
1.204.2.7  yamt      152: static struct pool_cache uvm_map_entry_cache;
1.130     thorpej   153:
                    154: MALLOC_DEFINE(M_VMMAP, "VM map", "VM map structures");
                    155: MALLOC_DEFINE(M_VMPMAP, "VM pmap", "VM pmap");
1.25      thorpej   156:
1.40      thorpej   157: #ifdef PMAP_GROWKERNEL
                    158: /*
                    159:  * This global represents the end of the kernel virtual address
                    160:  * space.  If we want to exceed this, we must grow the kernel
                    161:  * virtual address space dynamically.
                    162:  *
                    163:  * Note, this variable is locked by kernel_map's lock.
                    164:  */
                    165: vaddr_t uvm_maxkaddr;
                    166: #endif
                    167:
1.25      thorpej   168: /*
1.1       mrg       169:  * macros
                    170:  */
                    171:
                    172: /*
1.174     yamt      173:  * VM_MAP_USE_KMAPENT: determine if uvm_kmapent_alloc/free is used
                    174:  * for the vm_map.
                    175:  */
                    176: extern struct vm_map *pager_map; /* XXX */
1.204.2.1  yamt      177: #define        VM_MAP_USE_KMAPENT_FLAGS(flags) \
                    178:        (((flags) & VM_MAP_INTRSAFE) != 0)
1.174     yamt      179: #define        VM_MAP_USE_KMAPENT(map) \
1.204.2.1  yamt      180:        (VM_MAP_USE_KMAPENT_FLAGS((map)->flags) || (map) == kernel_map)
1.174     yamt      181:
                    182: /*
1.194     yamt      183:  * UVM_ET_ISCOMPATIBLE: check some requirements for map entry merging
                    184:  */
                    185:
                    186: #define        UVM_ET_ISCOMPATIBLE(ent, type, uobj, meflags, \
                    187:     prot, maxprot, inh, adv, wire) \
                    188:        ((ent)->etype == (type) && \
                    189:        (((ent)->flags ^ (meflags)) & (UVM_MAP_NOMERGE | UVM_MAP_QUANTUM)) \
                    190:        == 0 && \
                    191:        (ent)->object.uvm_obj == (uobj) && \
                    192:        (ent)->protection == (prot) && \
                    193:        (ent)->max_protection == (maxprot) && \
                    194:        (ent)->inheritance == (inh) && \
                    195:        (ent)->advice == (adv) && \
                    196:        (ent)->wired_count == (wire))
                    197:
                    198: /*
1.1       mrg       199:  * uvm_map_entry_link: insert entry into a map
                    200:  *
                    201:  * => map must be locked
                    202:  */
1.10      mrg       203: #define uvm_map_entry_link(map, after_where, entry) do { \
1.204.2.1  yamt      204:        uvm_mapent_check(entry); \
1.10      mrg       205:        (map)->nentries++; \
                    206:        (entry)->prev = (after_where); \
                    207:        (entry)->next = (after_where)->next; \
                    208:        (entry)->prev->next = (entry); \
                    209:        (entry)->next->prev = (entry); \
1.144     yamt      210:        uvm_rb_insert((map), (entry)); \
1.124     perry     211: } while (/*CONSTCOND*/ 0)
1.10      mrg       212:
1.1       mrg       213: /*
                    214:  * uvm_map_entry_unlink: remove entry from a map
                    215:  *
                    216:  * => map must be locked
                    217:  */
1.10      mrg       218: #define uvm_map_entry_unlink(map, entry) do { \
1.204.2.1  yamt      219:        KASSERT((entry) != (map)->first_free); \
                    220:        KASSERT((entry) != (map)->hint); \
                    221:        uvm_mapent_check(entry); \
1.10      mrg       222:        (map)->nentries--; \
                    223:        (entry)->next->prev = (entry)->prev; \
                    224:        (entry)->prev->next = (entry)->next; \
1.144     yamt      225:        uvm_rb_remove((map), (entry)); \
1.124     perry     226: } while (/*CONSTCOND*/ 0)
1.1       mrg       227:
                    228: /*
                    229:  * SAVE_HINT: saves the specified entry as the hint for future lookups.
                    230:  *
1.204.2.7  yamt      231:  * => map need not be locked.
1.1       mrg       232:  */
1.204.2.7  yamt      233: #define SAVE_HINT(map, check, value) do { \
                    234:        atomic_cas_ptr(&(map)->hint, (check), (value)); \
1.124     perry     235: } while (/*CONSTCOND*/ 0)
1.1       mrg       236:
                    237: /*
1.204.2.1  yamt      238:  * clear_hints: ensure that hints don't point to the entry.
                    239:  *
                    240:  * => map must be write-locked.
                    241:  */
                    242: static void
                    243: clear_hints(struct vm_map *map, struct vm_map_entry *ent)
                    244: {
                    245:
                    246:        SAVE_HINT(map, ent, ent->prev);
                    247:        if (map->first_free == ent) {
                    248:                map->first_free = ent->prev;
                    249:        }
                    250: }
                    251:
                    252: /*
1.1       mrg       253:  * VM_MAP_RANGE_CHECK: check and correct range
                    254:  *
                    255:  * => map must at least be read locked
                    256:  */
                    257:
1.10      mrg       258: #define VM_MAP_RANGE_CHECK(map, start, end) do { \
1.139     enami     259:        if (start < vm_map_min(map))            \
                    260:                start = vm_map_min(map);        \
                    261:        if (end > vm_map_max(map))              \
                    262:                end = vm_map_max(map);          \
                    263:        if (start > end)                        \
                    264:                start = end;                    \
1.124     perry     265: } while (/*CONSTCOND*/ 0)
1.1       mrg       266:
                    267: /*
                    268:  * local prototypes
                    269:  */
                    270:
1.138     enami     271: static struct vm_map_entry *
                    272:                uvm_mapent_alloc(struct vm_map *, int);
1.174     yamt      273: static struct vm_map_entry *
                    274:                uvm_mapent_alloc_split(struct vm_map *,
                    275:                    const struct vm_map_entry *, int,
                    276:                    struct uvm_mapent_reservation *);
1.138     enami     277: static void    uvm_mapent_copy(struct vm_map_entry *, struct vm_map_entry *);
                    278: static void    uvm_mapent_free(struct vm_map_entry *);
1.204.2.1  yamt      279: #if defined(DEBUG)
                    280: static void    _uvm_mapent_check(const struct vm_map_entry *, const char *,
                    281:                    int);
                    282: #define        uvm_mapent_check(map)   _uvm_mapent_check(map, __FILE__, __LINE__)
                    283: #else /* defined(DEBUG) */
                    284: #define        uvm_mapent_check(e)     /* nothing */
                    285: #endif /* defined(DEBUG) */
1.174     yamt      286: static struct vm_map_entry *
                    287:                uvm_kmapent_alloc(struct vm_map *, int);
                    288: static void    uvm_kmapent_free(struct vm_map_entry *);
1.204.2.1  yamt      289: static vsize_t uvm_kmapent_overhead(vsize_t);
                    290:
1.138     enami     291: static void    uvm_map_entry_unwire(struct vm_map *, struct vm_map_entry *);
                    292: static void    uvm_map_reference_amap(struct vm_map_entry *, int);
1.140     enami     293: static int     uvm_map_space_avail(vaddr_t *, vsize_t, voff_t, vsize_t, int,
                    294:                    struct vm_map_entry *);
1.138     enami     295: static void    uvm_map_unreference_amap(struct vm_map_entry *, int);
1.1       mrg       296:
1.204.2.1  yamt      297: int _uvm_map_sanity(struct vm_map *);
                    298: int _uvm_tree_sanity(struct vm_map *);
1.144     yamt      299: static vsize_t uvm_rb_subtree_space(const struct vm_map_entry *);
                    300:
1.204.2.1  yamt      301: static inline int
1.144     yamt      302: uvm_compare(const struct vm_map_entry *a, const struct vm_map_entry *b)
                    303: {
                    304:
                    305:        if (a->start < b->start)
                    306:                return (-1);
                    307:        else if (a->start > b->start)
                    308:                return (1);
1.164     junyoung  309:
1.144     yamt      310:        return (0);
                    311: }
                    312:
1.204.2.1  yamt      313: static inline void
1.144     yamt      314: uvm_rb_augment(struct vm_map_entry *entry)
                    315: {
                    316:
                    317:        entry->space = uvm_rb_subtree_space(entry);
                    318: }
                    319:
                    320: RB_PROTOTYPE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
                    321:
                    322: RB_GENERATE(uvm_tree, vm_map_entry, rb_entry, uvm_compare);
                    323:
1.204.2.1  yamt      324: static inline vsize_t
1.144     yamt      325: uvm_rb_space(const struct vm_map *map, const struct vm_map_entry *entry)
                    326: {
                    327:        /* XXX map is not used */
                    328:
                    329:        KASSERT(entry->next != NULL);
                    330:        return entry->next->start - entry->end;
                    331: }
                    332:
                    333: static vsize_t
                    334: uvm_rb_subtree_space(const struct vm_map_entry *entry)
                    335: {
                    336:        vaddr_t space, tmp;
                    337:
                    338:        space = entry->ownspace;
                    339:        if (RB_LEFT(entry, rb_entry)) {
                    340:                tmp = RB_LEFT(entry, rb_entry)->space;
                    341:                if (tmp > space)
                    342:                        space = tmp;
                    343:        }
                    344:
                    345:        if (RB_RIGHT(entry, rb_entry)) {
                    346:                tmp = RB_RIGHT(entry, rb_entry)->space;
                    347:                if (tmp > space)
                    348:                        space = tmp;
                    349:        }
                    350:
                    351:        return (space);
                    352: }
                    353:
1.204.2.1  yamt      354: static inline void
1.144     yamt      355: uvm_rb_fixup(struct vm_map *map, struct vm_map_entry *entry)
                    356: {
                    357:        /* We need to traverse to the very top */
                    358:        do {
                    359:                entry->ownspace = uvm_rb_space(map, entry);
                    360:                entry->space = uvm_rb_subtree_space(entry);
                    361:        } while ((entry = RB_PARENT(entry, rb_entry)) != NULL);
                    362: }
                    363:
1.203     thorpej   364: static void
1.144     yamt      365: uvm_rb_insert(struct vm_map *map, struct vm_map_entry *entry)
                    366: {
                    367:        vaddr_t space = uvm_rb_space(map, entry);
                    368:        struct vm_map_entry *tmp;
                    369:
                    370:        entry->ownspace = entry->space = space;
                    371:        tmp = RB_INSERT(uvm_tree, &(map)->rbhead, entry);
                    372: #ifdef DIAGNOSTIC
                    373:        if (tmp != NULL)
                    374:                panic("uvm_rb_insert: duplicate entry?");
                    375: #endif
                    376:        uvm_rb_fixup(map, entry);
                    377:        if (entry->prev != &map->header)
                    378:                uvm_rb_fixup(map, entry->prev);
                    379: }
                    380:
1.203     thorpej   381: static void
1.144     yamt      382: uvm_rb_remove(struct vm_map *map, struct vm_map_entry *entry)
                    383: {
                    384:        struct vm_map_entry *parent;
                    385:
                    386:        parent = RB_PARENT(entry, rb_entry);
                    387:        RB_REMOVE(uvm_tree, &(map)->rbhead, entry);
                    388:        if (entry->prev != &map->header)
                    389:                uvm_rb_fixup(map, entry->prev);
                    390:        if (parent)
                    391:                uvm_rb_fixup(map, parent);
                    392: }
                    393:
1.204.2.1  yamt      394: #if defined(DEBUG)
                    395: int uvm_debug_check_map = 0;
1.159     yamt      396: int uvm_debug_check_rbtree = 0;
1.204.2.1  yamt      397: #define uvm_map_check(map, name) \
                    398:        _uvm_map_check((map), (name), __FILE__, __LINE__)
                    399: static void
                    400: _uvm_map_check(struct vm_map *map, const char *name,
                    401:     const char *file, int line)
                    402: {
1.144     yamt      403:
1.204.2.1  yamt      404:        if ((uvm_debug_check_map && _uvm_map_sanity(map)) ||
                    405:            (uvm_debug_check_rbtree && _uvm_tree_sanity(map))) {
                    406:                panic("uvm_map_check failed: \"%s\" map=%p (%s:%d)",
                    407:                    name, map, file, line);
                    408:        }
                    409: }
                    410: #else /* defined(DEBUG) */
                    411: #define uvm_map_check(map, name)       /* nothing */
                    412: #endif /* defined(DEBUG) */
                    413:
                    414: #if defined(DEBUG) || defined(DDB)
1.144     yamt      415: int
1.204.2.1  yamt      416: _uvm_map_sanity(struct vm_map *map)
                    417: {
1.204.2.3  yamt      418:        bool first_free_found = false;
                    419:        bool hint_found = false;
1.204.2.1  yamt      420:        const struct vm_map_entry *e;
                    421:
                    422:        e = &map->header;
                    423:        for (;;) {
                    424:                if (map->first_free == e) {
1.204.2.3  yamt      425:                        first_free_found = true;
1.204.2.1  yamt      426:                } else if (!first_free_found && e->next->start > e->end) {
                    427:                        printf("first_free %p should be %p\n",
                    428:                            map->first_free, e);
                    429:                        return -1;
                    430:                }
                    431:                if (map->hint == e) {
1.204.2.3  yamt      432:                        hint_found = true;
1.204.2.1  yamt      433:                }
                    434:
                    435:                e = e->next;
                    436:                if (e == &map->header) {
                    437:                        break;
                    438:                }
                    439:        }
                    440:        if (!first_free_found) {
                    441:                printf("stale first_free\n");
                    442:                return -1;
                    443:        }
                    444:        if (!hint_found) {
                    445:                printf("stale hint\n");
                    446:                return -1;
                    447:        }
                    448:        return 0;
                    449: }
                    450:
                    451: int
                    452: _uvm_tree_sanity(struct vm_map *map)
1.144     yamt      453: {
                    454:        struct vm_map_entry *tmp, *trtmp;
                    455:        int n = 0, i = 1;
                    456:
                    457:        RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
                    458:                if (tmp->ownspace != uvm_rb_space(map, tmp)) {
1.204.2.1  yamt      459:                        printf("%d/%d ownspace %lx != %lx %s\n",
                    460:                            n + 1, map->nentries,
1.144     yamt      461:                            (ulong)tmp->ownspace, (ulong)uvm_rb_space(map, tmp),
                    462:                            tmp->next == &map->header ? "(last)" : "");
                    463:                        goto error;
                    464:                }
                    465:        }
                    466:        trtmp = NULL;
                    467:        RB_FOREACH(tmp, uvm_tree, &map->rbhead) {
                    468:                if (tmp->space != uvm_rb_subtree_space(tmp)) {
1.204.2.1  yamt      469:                        printf("space %lx != %lx\n",
                    470:                            (ulong)tmp->space,
1.144     yamt      471:                            (ulong)uvm_rb_subtree_space(tmp));
                    472:                        goto error;
                    473:                }
                    474:                if (trtmp != NULL && trtmp->start >= tmp->start) {
1.204.2.1  yamt      475:                        printf("corrupt: 0x%lx >= 0x%lx\n",
                    476:                            trtmp->start, tmp->start);
1.144     yamt      477:                        goto error;
                    478:                }
                    479:                n++;
                    480:
                    481:                trtmp = tmp;
                    482:        }
                    483:
                    484:        if (n != map->nentries) {
1.204.2.1  yamt      485:                printf("nentries: %d vs %d\n", n, map->nentries);
1.144     yamt      486:                goto error;
                    487:        }
                    488:
                    489:        for (tmp = map->header.next; tmp && tmp != &map->header;
                    490:            tmp = tmp->next, i++) {
                    491:                trtmp = RB_FIND(uvm_tree, &map->rbhead, tmp);
                    492:                if (trtmp != tmp) {
1.204.2.1  yamt      493:                        printf("lookup: %d: %p - %p: %p\n", i, tmp, trtmp,
1.144     yamt      494:                            RB_PARENT(tmp, rb_entry));
                    495:                        goto error;
                    496:                }
                    497:        }
                    498:
                    499:        return (0);
                    500:  error:
                    501:        return (-1);
                    502: }
1.204.2.1  yamt      503: #endif /* defined(DEBUG) || defined(DDB) */
1.144     yamt      504:
1.201     dsl       505: #ifdef DIAGNOSTIC
1.203     thorpej   506: static struct vm_map *uvm_kmapent_map(struct vm_map_entry *);
1.201     dsl       507: #endif
1.174     yamt      508:
1.1       mrg       509: /*
1.204.2.4  yamt      510:  * vm_map_lock: acquire an exclusive (write) lock on a map.
                    511:  *
                    512:  * => Note that "intrsafe" maps use only exclusive, spin locks.
                    513:  *
                    514:  * => The locking protocol provides for guaranteed upgrade from shared ->
                    515:  *    exclusive by whichever thread currently has the map marked busy.
                    516:  *    See "LOCKING PROTOCOL NOTES" in uvm_map.h.  This is horrible; among
                    517:  *    other problems, it defeats any fairness guarantees provided by RW
                    518:  *    locks.
                    519:  */
                    520:
                    521: void
                    522: vm_map_lock(struct vm_map *map)
                    523: {
                    524:
                    525:        if ((map->flags & VM_MAP_INTRSAFE) != 0) {
                    526:                mutex_spin_enter(&map->mutex);
                    527:                return;
                    528:        }
                    529:
                    530:        for (;;) {
                    531:                rw_enter(&map->lock, RW_WRITER);
                    532:                if (map->busy == NULL)
                    533:                        break;
1.204.2.7  yamt      534:                if (map->busy == curlwp)
                    535:                        break;
1.204.2.4  yamt      536:                mutex_enter(&map->misc_lock);
                    537:                rw_exit(&map->lock);
1.204.2.7  yamt      538:                if (map->busy != NULL)
                    539:                        cv_wait(&map->cv, &map->misc_lock);
1.204.2.4  yamt      540:                mutex_exit(&map->misc_lock);
                    541:        }
                    542:
                    543:        map->timestamp++;
                    544: }
                    545:
                    546: /*
                    547:  * vm_map_lock_try: try to lock a map, failing if it is already locked.
                    548:  */
                    549:
                    550: bool
                    551: vm_map_lock_try(struct vm_map *map)
                    552: {
                    553:
                    554:        if ((map->flags & VM_MAP_INTRSAFE) != 0)
                    555:                return mutex_tryenter(&map->mutex);
                    556:        if (!rw_tryenter(&map->lock, RW_WRITER))
                    557:                return false;
                    558:        if (map->busy != NULL) {
                    559:                rw_exit(&map->lock);
                    560:                return false;
                    561:        }
                    562:
                    563:        map->timestamp++;
                    564:        return true;
                    565: }
                    566:
                    567: /*
                    568:  * vm_map_unlock: release an exclusive lock on a map.
                    569:  */
                    570:
                    571: void
                    572: vm_map_unlock(struct vm_map *map)
                    573: {
                    574:
                    575:        if ((map->flags & VM_MAP_INTRSAFE) != 0)
                    576:                mutex_spin_exit(&map->mutex);
                    577:        else {
                    578:                KASSERT(rw_write_held(&map->lock));
1.204.2.7  yamt      579:                KASSERT(map->busy == NULL || map->busy == curlwp);
1.204.2.4  yamt      580:                rw_exit(&map->lock);
                    581:        }
                    582: }
                    583:
                    584: /*
                    585:  * vm_map_unbusy: mark the map as unbusy, and wake any waiters that
                    586:  *     want an exclusive lock.
                    587:  */
                    588:
                    589: void
                    590: vm_map_unbusy(struct vm_map *map)
                    591: {
                    592:
                    593:        KASSERT(map->busy == curlwp);
                    594:
                    595:        /*
                    596:         * Safe to clear 'busy' and 'waiters' with only a read lock held:
                    597:         *
                    598:         * o they can only be set with a write lock held
                    599:         * o writers are blocked out with a read or write hold
                    600:         * o at any time, only one thread owns the set of values
                    601:         */
                    602:        mutex_enter(&map->misc_lock);
1.204.2.7  yamt      603:        map->busy = NULL;
1.204.2.4  yamt      604:        cv_broadcast(&map->cv);
                    605:        mutex_exit(&map->misc_lock);
                    606: }
                    607:
                    608: /*
1.204.2.7  yamt      609:  * vm_map_lock_read: acquire a shared (read) lock on a map.
                    610:  */
                    611:
                    612: void
                    613: vm_map_lock_read(struct vm_map *map)
                    614: {
                    615:
                    616:        KASSERT((map->flags & VM_MAP_INTRSAFE) == 0);
                    617:
                    618:        rw_enter(&map->lock, RW_READER);
                    619: }
                    620:
                    621: /*
                    622:  * vm_map_unlock_read: release a shared lock on a map.
                    623:  */
                    624:
                    625: void
                    626: vm_map_unlock_read(struct vm_map *map)
                    627: {
                    628:
                    629:        KASSERT((map->flags & VM_MAP_INTRSAFE) == 0);
                    630:
                    631:        rw_exit(&map->lock);
                    632: }
                    633:
                    634: /*
                    635:  * vm_map_busy: mark a map as busy.
                    636:  *
                    637:  * => the caller must hold the map write locked
                    638:  */
                    639:
                    640: void
                    641: vm_map_busy(struct vm_map *map)
                    642: {
                    643:
                    644:        KASSERT(rw_write_held(&map->lock));
                    645:        KASSERT(map->busy == NULL);
                    646:
                    647:        map->busy = curlwp;
                    648: }
                    649:
                    650: /*
                    651:  * vm_map_locked_p: return true if the map is write locked.
                    652:  */
                    653:
                    654: bool
                    655: vm_map_locked_p(struct vm_map *map)
                    656: {
                    657:
                    658:        if ((map->flags & VM_MAP_INTRSAFE) != 0) {
                    659:                return mutex_owned(&map->mutex);
                    660:        } else {
                    661:                return rw_write_held(&map->lock);
                    662:        }
                    663: }
                    664:
                    665: /*
1.1       mrg       666:  * uvm_mapent_alloc: allocate a map entry
                    667:  */
                    668:
1.203     thorpej   669: static struct vm_map_entry *
1.138     enami     670: uvm_mapent_alloc(struct vm_map *map, int flags)
1.10      mrg       671: {
1.99      chs       672:        struct vm_map_entry *me;
1.127     thorpej   673:        int pflags = (flags & UVM_FLAG_NOWAIT) ? PR_NOWAIT : PR_WAITOK;
1.104     chs       674:        UVMHIST_FUNC("uvm_mapent_alloc"); UVMHIST_CALLED(maphist);
1.1       mrg       675:
1.174     yamt      676:        if (VM_MAP_USE_KMAPENT(map)) {
                    677:                me = uvm_kmapent_alloc(map, flags);
1.104     chs       678:        } else {
1.204.2.7  yamt      679:                me = pool_cache_get(&uvm_map_entry_cache, pflags);
1.126     bouyer    680:                if (__predict_false(me == NULL))
                    681:                        return NULL;
1.104     chs       682:                me->flags = 0;
1.10      mrg       683:        }
1.1       mrg       684:
1.104     chs       685:        UVMHIST_LOG(maphist, "<- new entry=0x%x [kentry=%d]", me,
                    686:            ((map->flags & VM_MAP_INTRSAFE) != 0 || map == kernel_map), 0, 0);
1.139     enami     687:        return (me);
1.1       mrg       688: }
                    689:
                    690: /*
1.174     yamt      691:  * uvm_mapent_alloc_split: allocate a map entry for clipping.
1.204.2.7  yamt      692:  *
                    693:  * => map must be locked by caller if UVM_MAP_QUANTUM is set.
1.174     yamt      694:  */
                    695:
1.203     thorpej   696: static struct vm_map_entry *
1.174     yamt      697: uvm_mapent_alloc_split(struct vm_map *map,
                    698:     const struct vm_map_entry *old_entry, int flags,
                    699:     struct uvm_mapent_reservation *umr)
                    700: {
                    701:        struct vm_map_entry *me;
                    702:
                    703:        KASSERT(!VM_MAP_USE_KMAPENT(map) ||
                    704:            (old_entry->flags & UVM_MAP_QUANTUM) || !UMR_EMPTY(umr));
                    705:
                    706:        if (old_entry->flags & UVM_MAP_QUANTUM) {
1.175     yamt      707:                struct vm_map_kernel *vmk = vm_map_to_kernel(map);
1.174     yamt      708:
1.204.2.7  yamt      709:                KASSERT(vm_map_locked_p(map));
1.175     yamt      710:                me = vmk->vmk_merged_entries;
1.174     yamt      711:                KASSERT(me);
1.175     yamt      712:                vmk->vmk_merged_entries = me->next;
1.174     yamt      713:                KASSERT(me->flags & UVM_MAP_QUANTUM);
                    714:        } else {
                    715:                me = uvm_mapent_alloc(map, flags);
                    716:        }
                    717:
                    718:        return me;
                    719: }
                    720:
                    721: /*
1.1       mrg       722:  * uvm_mapent_free: free map entry
                    723:  */
                    724:
1.203     thorpej   725: static void
1.138     enami     726: uvm_mapent_free(struct vm_map_entry *me)
1.1       mrg       727: {
1.104     chs       728:        UVMHIST_FUNC("uvm_mapent_free"); UVMHIST_CALLED(maphist);
                    729:
1.98      chs       730:        UVMHIST_LOG(maphist,"<- freeing map entry=0x%x [flags=%d]",
1.1       mrg       731:                me, me->flags, 0, 0);
1.174     yamt      732:        if (me->flags & UVM_MAP_KERNEL) {
                    733:                uvm_kmapent_free(me);
                    734:        } else {
1.204.2.7  yamt      735:                pool_cache_put(&uvm_map_entry_cache, me);
1.174     yamt      736:        }
                    737: }
                    738:
                    739: /*
1.194     yamt      740:  * uvm_mapent_free_merged: free merged map entry
1.174     yamt      741:  *
                    742:  * => keep the entry if needed.
1.194     yamt      743:  * => caller shouldn't hold map locked if VM_MAP_USE_KMAPENT(map) is true.
1.204.2.7  yamt      744:  * => map should be locked if UVM_MAP_QUANTUM is set.
1.174     yamt      745:  */
                    746:
1.203     thorpej   747: static void
1.181     yamt      748: uvm_mapent_free_merged(struct vm_map *map, struct vm_map_entry *me)
1.174     yamt      749: {
                    750:
1.181     yamt      751:        KASSERT(!(me->flags & UVM_MAP_KERNEL) || uvm_kmapent_map(me) == map);
                    752:
1.174     yamt      753:        if (me->flags & UVM_MAP_QUANTUM) {
                    754:                /*
                    755:                 * keep this entry for later splitting.
                    756:                 */
1.175     yamt      757:                struct vm_map_kernel *vmk;
1.174     yamt      758:
1.204.2.7  yamt      759:                KASSERT(vm_map_locked_p(map));
1.181     yamt      760:                KASSERT(VM_MAP_IS_KERNEL(map));
                    761:                KASSERT(!VM_MAP_USE_KMAPENT(map) ||
                    762:                    (me->flags & UVM_MAP_KERNEL));
1.174     yamt      763:
1.175     yamt      764:                vmk = vm_map_to_kernel(map);
                    765:                me->next = vmk->vmk_merged_entries;
                    766:                vmk->vmk_merged_entries = me;
1.104     chs       767:        } else {
1.174     yamt      768:                uvm_mapent_free(me);
1.10      mrg       769:        }
1.1       mrg       770: }
                    771:
                    772: /*
                    773:  * uvm_mapent_copy: copy a map entry, preserving flags
                    774:  */
                    775:
1.204.2.1  yamt      776: static inline void
1.138     enami     777: uvm_mapent_copy(struct vm_map_entry *src, struct vm_map_entry *dst)
1.10      mrg       778: {
1.139     enami     779:
1.106     chs       780:        memcpy(dst, src, ((char *)&src->uvm_map_entry_stop_copy) -
1.139     enami     781:            ((char *)src));
1.1       mrg       782: }
                    783:
                    784: /*
1.204.2.1  yamt      785:  * uvm_mapent_overhead: calculate maximum kva overhead necessary for
                    786:  * map entries.
                    787:  *
                    788:  * => size and flags are the same as uvm_km_suballoc's ones.
                    789:  */
                    790:
                    791: vsize_t
                    792: uvm_mapent_overhead(vsize_t size, int flags)
                    793: {
                    794:
                    795:        if (VM_MAP_USE_KMAPENT_FLAGS(flags)) {
                    796:                return uvm_kmapent_overhead(size);
                    797:        }
                    798:        return 0;
                    799: }
                    800:
                    801: #if defined(DEBUG)
                    802: static void
                    803: _uvm_mapent_check(const struct vm_map_entry *entry, const char *file, int line)
                    804: {
                    805:
                    806:        if (entry->start >= entry->end) {
                    807:                goto bad;
                    808:        }
                    809:        if (UVM_ET_ISOBJ(entry)) {
                    810:                if (entry->object.uvm_obj == NULL) {
                    811:                        goto bad;
                    812:                }
                    813:        } else if (UVM_ET_ISSUBMAP(entry)) {
                    814:                if (entry->object.sub_map == NULL) {
                    815:                        goto bad;
                    816:                }
                    817:        } else {
                    818:                if (entry->object.uvm_obj != NULL ||
                    819:                    entry->object.sub_map != NULL) {
                    820:                        goto bad;
                    821:                }
                    822:        }
                    823:        if (!UVM_ET_ISOBJ(entry)) {
                    824:                if (entry->offset != 0) {
                    825:                        goto bad;
                    826:                }
                    827:        }
                    828:
                    829:        return;
                    830:
                    831: bad:
                    832:        panic("%s: bad entry %p (%s:%d)", __func__, entry, file, line);
                    833: }
                    834: #endif /* defined(DEBUG) */
                    835:
                    836: /*
1.1       mrg       837:  * uvm_map_entry_unwire: unwire a map entry
                    838:  *
                    839:  * => map should be locked by caller
                    840:  */
                    841:
1.204.2.1  yamt      842: static inline void
1.138     enami     843: uvm_map_entry_unwire(struct vm_map *map, struct vm_map_entry *entry)
1.10      mrg       844: {
1.139     enami     845:
1.10      mrg       846:        entry->wired_count = 0;
1.57      thorpej   847:        uvm_fault_unwire_locked(map, entry->start, entry->end);
1.1       mrg       848: }
                    849:
1.85      chs       850:
                    851: /*
                    852:  * wrapper for calling amap_ref()
                    853:  */
1.204.2.1  yamt      854: static inline void
1.138     enami     855: uvm_map_reference_amap(struct vm_map_entry *entry, int flags)
1.85      chs       856: {
1.139     enami     857:
1.99      chs       858:        amap_ref(entry->aref.ar_amap, entry->aref.ar_pageoff,
1.139     enami     859:            (entry->end - entry->start) >> PAGE_SHIFT, flags);
1.85      chs       860: }
                    861:
                    862:
                    863: /*
1.98      chs       864:  * wrapper for calling amap_unref()
1.85      chs       865:  */
1.204.2.1  yamt      866: static inline void
1.138     enami     867: uvm_map_unreference_amap(struct vm_map_entry *entry, int flags)
1.85      chs       868: {
1.139     enami     869:
1.99      chs       870:        amap_unref(entry->aref.ar_amap, entry->aref.ar_pageoff,
1.139     enami     871:            (entry->end - entry->start) >> PAGE_SHIFT, flags);
1.85      chs       872: }
                    873:
                    874:
1.1       mrg       875: /*
1.204.2.7  yamt      876:  * uvm_map_init: init mapping system at boot time.
1.1       mrg       877:  */
                    878:
1.10      mrg       879: void
1.138     enami     880: uvm_map_init(void)
1.1       mrg       881: {
                    882: #if defined(UVMHIST)
1.10      mrg       883:        static struct uvm_history_ent maphistbuf[100];
                    884:        static struct uvm_history_ent pdhistbuf[100];
1.1       mrg       885: #endif
1.10      mrg       886:
                    887:        /*
                    888:         * first, init logging system.
                    889:         */
1.1       mrg       890:
1.10      mrg       891:        UVMHIST_FUNC("uvm_map_init");
                    892:        UVMHIST_INIT_STATIC(maphist, maphistbuf);
                    893:        UVMHIST_INIT_STATIC(pdhist, pdhistbuf);
                    894:        UVMHIST_CALLED(maphist);
                    895:        UVMHIST_LOG(maphist,"<starting uvm map system>", 0, 0, 0, 0);
                    896:
                    897:        /*
1.174     yamt      898:         * initialize the global lock for kernel map entry.
1.10      mrg       899:         */
                    900:
1.204.2.4  yamt      901:        mutex_init(&uvm_kentry_lock, MUTEX_DRIVER, IPL_VM);
1.204.2.7  yamt      902:
                    903:        /*
                    904:         * initialize caches.
                    905:         */
                    906:
                    907:        pool_cache_bootstrap(&uvm_map_entry_cache, sizeof(struct vm_map_entry),
                    908:            0, 0, 0, "vmmpepl", NULL, IPL_NONE, NULL, NULL, NULL);
                    909:        pool_cache_bootstrap(&uvm_vmspace_cache, sizeof(struct vmspace),
                    910:            0, 0, 0, "vmsppl", NULL, IPL_NONE, NULL, NULL, NULL);
1.1       mrg       911: }
                    912:
                    913: /*
                    914:  * clippers
                    915:  */
                    916:
                    917: /*
1.204.2.1  yamt      918:  * uvm_mapent_splitadj: adjust map entries for splitting, after uvm_mapent_copy.
                    919:  */
                    920:
                    921: static void
                    922: uvm_mapent_splitadj(struct vm_map_entry *entry1, struct vm_map_entry *entry2,
                    923:     vaddr_t splitat)
                    924: {
                    925:        vaddr_t adj;
                    926:
                    927:        KASSERT(entry1->start < splitat);
                    928:        KASSERT(splitat < entry1->end);
                    929:
                    930:        adj = splitat - entry1->start;
                    931:        entry1->end = entry2->start = splitat;
                    932:
                    933:        if (entry1->aref.ar_amap) {
                    934:                amap_splitref(&entry1->aref, &entry2->aref, adj);
                    935:        }
                    936:        if (UVM_ET_ISSUBMAP(entry1)) {
                    937:                /* ... unlikely to happen, but play it safe */
                    938:                 uvm_map_reference(entry1->object.sub_map);
                    939:        } else if (UVM_ET_ISOBJ(entry1)) {
                    940:                KASSERT(entry1->object.uvm_obj != NULL); /* suppress coverity */
                    941:                entry2->offset += adj;
                    942:                if (entry1->object.uvm_obj->pgops &&
                    943:                    entry1->object.uvm_obj->pgops->pgo_reference)
                    944:                        entry1->object.uvm_obj->pgops->pgo_reference(
                    945:                            entry1->object.uvm_obj);
                    946:        }
                    947: }
                    948:
                    949: /*
1.1       mrg       950:  * uvm_map_clip_start: ensure that the entry begins at or after
                    951:  *     the starting address, if it doesn't we split the entry.
1.98      chs       952:  *
1.1       mrg       953:  * => caller should use UVM_MAP_CLIP_START macro rather than calling
                    954:  *    this directly
                    955:  * => map must be locked by caller
                    956:  */
                    957:
1.99      chs       958: void
1.138     enami     959: uvm_map_clip_start(struct vm_map *map, struct vm_map_entry *entry,
1.174     yamt      960:     vaddr_t start, struct uvm_mapent_reservation *umr)
1.1       mrg       961: {
1.99      chs       962:        struct vm_map_entry *new_entry;
1.1       mrg       963:
                    964:        /* uvm_map_simplify_entry(map, entry); */ /* XXX */
                    965:
1.204.2.1  yamt      966:        uvm_map_check(map, "clip_start entry");
                    967:        uvm_mapent_check(entry);
1.144     yamt      968:
1.10      mrg       969:        /*
                    970:         * Split off the front portion.  note that we must insert the new
                    971:         * entry BEFORE this one, so that this entry has the specified
1.1       mrg       972:         * starting address.
1.10      mrg       973:         */
1.174     yamt      974:        new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
1.1       mrg       975:        uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
1.204.2.1  yamt      976:        uvm_mapent_splitadj(new_entry, entry, start);
1.10      mrg       977:        uvm_map_entry_link(map, entry->prev, new_entry);
1.85      chs       978:
1.204.2.1  yamt      979:        uvm_map_check(map, "clip_start leave");
1.1       mrg       980: }
                    981:
                    982: /*
                    983:  * uvm_map_clip_end: ensure that the entry ends at or before
                    984:  *     the ending address, if it does't we split the reference
1.98      chs       985:  *
1.1       mrg       986:  * => caller should use UVM_MAP_CLIP_END macro rather than calling
                    987:  *    this directly
                    988:  * => map must be locked by caller
                    989:  */
                    990:
1.10      mrg       991: void
1.174     yamt      992: uvm_map_clip_end(struct vm_map *map, struct vm_map_entry *entry, vaddr_t end,
                    993:     struct uvm_mapent_reservation *umr)
1.1       mrg       994: {
1.204.2.1  yamt      995:        struct vm_map_entry *new_entry;
1.1       mrg       996:
1.204.2.1  yamt      997:        uvm_map_check(map, "clip_end entry");
                    998:        uvm_mapent_check(entry);
1.174     yamt      999:
1.1       mrg      1000:        /*
                   1001:         *      Create a new entry and insert it
                   1002:         *      AFTER the specified entry
                   1003:         */
1.174     yamt     1004:        new_entry = uvm_mapent_alloc_split(map, entry, 0, umr);
1.1       mrg      1005:        uvm_mapent_copy(entry, new_entry); /* entry -> new_entry */
1.204.2.1  yamt     1006:        uvm_mapent_splitadj(entry, new_entry, end);
                   1007:        uvm_map_entry_link(map, entry, new_entry);
1.1       mrg      1008:
1.204.2.1  yamt     1009:        uvm_map_check(map, "clip_end leave");
                   1010: }
1.144     yamt     1011:
1.204.2.1  yamt     1012: static void
                   1013: vm_map_drain(struct vm_map *map, uvm_flag_t flags)
                   1014: {
1.1       mrg      1015:
1.204.2.1  yamt     1016:        if (!VM_MAP_IS_KERNEL(map)) {
                   1017:                return;
1.1       mrg      1018:        }
1.144     yamt     1019:
1.204.2.1  yamt     1020:        uvm_km_va_drain(map, flags);
1.1       mrg      1021: }
                   1022:
                   1023: /*
                   1024:  *   M A P   -   m a i n   e n t r y   p o i n t
                   1025:  */
                   1026: /*
                   1027:  * uvm_map: establish a valid mapping in a map
                   1028:  *
                   1029:  * => assume startp is page aligned.
                   1030:  * => assume size is a multiple of PAGE_SIZE.
                   1031:  * => assume sys_mmap provides enough of a "hint" to have us skip
                   1032:  *     over text/data/bss area.
                   1033:  * => map must be unlocked (we will lock it)
                   1034:  * => <uobj,uoffset> value meanings (4 cases):
1.139     enami    1035:  *      [1] <NULL,uoffset>             == uoffset is a hint for PMAP_PREFER
1.1       mrg      1036:  *      [2] <NULL,UVM_UNKNOWN_OFFSET>  == don't PMAP_PREFER
                   1037:  *      [3] <uobj,uoffset>             == normal mapping
                   1038:  *      [4] <uobj,UVM_UNKNOWN_OFFSET>  == uvm_map finds offset based on VA
1.98      chs      1039:  *
1.1       mrg      1040:  *    case [4] is for kernel mappings where we don't know the offset until
1.8       chuck    1041:  *    we've found a virtual address.   note that kernel object offsets are
                   1042:  *    always relative to vm_map_min(kernel_map).
1.81      thorpej  1043:  *
1.165     yamt     1044:  * => if `align' is non-zero, we align the virtual address to the specified
                   1045:  *     alignment.
                   1046:  *     this is provided as a mechanism for large pages.
1.81      thorpej  1047:  *
1.1       mrg      1048:  * => XXXCDC: need way to map in external amap?
                   1049:  */
                   1050:
1.10      mrg      1051: int
1.138     enami    1052: uvm_map(struct vm_map *map, vaddr_t *startp /* IN/OUT */, vsize_t size,
                   1053:     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags)
1.10      mrg      1054: {
1.174     yamt     1055:        struct uvm_map_args args;
                   1056:        struct vm_map_entry *new_entry;
                   1057:        int error;
                   1058:
1.181     yamt     1059:        KASSERT((flags & UVM_FLAG_QUANTUM) == 0 || VM_MAP_IS_KERNEL(map));
1.187     yamt     1060:        KASSERT((size & PAGE_MASK) == 0);
1.174     yamt     1061:
                   1062:        /*
                   1063:         * for pager_map, allocate the new entry first to avoid sleeping
                   1064:         * for memory while we have the map locked.
                   1065:         *
                   1066:         * besides, because we allocates entries for in-kernel maps
                   1067:         * a bit differently (cf. uvm_kmapent_alloc/free), we need to
                   1068:         * allocate them before locking the map.
                   1069:         */
                   1070:
                   1071:        new_entry = NULL;
1.181     yamt     1072:        if (VM_MAP_USE_KMAPENT(map) || (flags & UVM_FLAG_QUANTUM) ||
                   1073:            map == pager_map) {
1.174     yamt     1074:                new_entry = uvm_mapent_alloc(map, (flags & UVM_FLAG_NOWAIT));
                   1075:                if (__predict_false(new_entry == NULL))
                   1076:                        return ENOMEM;
                   1077:                if (flags & UVM_FLAG_QUANTUM)
                   1078:                        new_entry->flags |= UVM_MAP_QUANTUM;
                   1079:        }
                   1080:        if (map == pager_map)
                   1081:                flags |= UVM_FLAG_NOMERGE;
                   1082:
                   1083:        error = uvm_map_prepare(map, *startp, size, uobj, uoffset, align,
                   1084:            flags, &args);
                   1085:        if (!error) {
                   1086:                error = uvm_map_enter(map, &args, new_entry);
                   1087:                *startp = args.uma_start;
1.189     yamt     1088:        } else if (new_entry) {
                   1089:                uvm_mapent_free(new_entry);
1.174     yamt     1090:        }
                   1091:
1.187     yamt     1092: #if defined(DEBUG)
                   1093:        if (!error && VM_MAP_IS_KERNEL(map)) {
                   1094:                uvm_km_check_empty(*startp, *startp + size,
                   1095:                    (map->flags & VM_MAP_INTRSAFE) != 0);
                   1096:        }
                   1097: #endif /* defined(DEBUG) */
                   1098:
1.174     yamt     1099:        return error;
                   1100: }
                   1101:
                   1102: int
                   1103: uvm_map_prepare(struct vm_map *map, vaddr_t start, vsize_t size,
                   1104:     struct uvm_object *uobj, voff_t uoffset, vsize_t align, uvm_flag_t flags,
                   1105:     struct uvm_map_args *args)
                   1106: {
                   1107:        struct vm_map_entry *prev_entry;
                   1108:        vm_prot_t prot = UVM_PROTECTION(flags);
                   1109:        vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
                   1110:
                   1111:        UVMHIST_FUNC("uvm_map_prepare");
1.10      mrg      1112:        UVMHIST_CALLED(maphist);
1.1       mrg      1113:
1.174     yamt     1114:        UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
                   1115:            map, start, size, flags);
1.10      mrg      1116:        UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
1.107     chs      1117:
                   1118:        /*
                   1119:         * detect a popular device driver bug.
                   1120:         */
                   1121:
1.139     enami    1122:        KASSERT(doing_shutdown || curlwp != NULL ||
1.129     christos 1123:            (map->flags & VM_MAP_INTRSAFE));
1.1       mrg      1124:
1.10      mrg      1125:        /*
1.144     yamt     1126:         * zero-sized mapping doesn't make any sense.
                   1127:         */
                   1128:        KASSERT(size > 0);
                   1129:
1.180     yamt     1130:        KASSERT((~flags & (UVM_FLAG_NOWAIT | UVM_FLAG_WAITVA)) != 0);
                   1131:
1.204.2.1  yamt     1132:        uvm_map_check(map, "map entry");
1.144     yamt     1133:
                   1134:        /*
1.106     chs      1135:         * check sanity of protection code
1.10      mrg      1136:         */
1.1       mrg      1137:
1.10      mrg      1138:        if ((prot & maxprot) != prot) {
1.98      chs      1139:                UVMHIST_LOG(maphist, "<- prot. failure:  prot=0x%x, max=0x%x",
1.10      mrg      1140:                prot, maxprot,0,0);
1.94      chs      1141:                return EACCES;
1.10      mrg      1142:        }
1.1       mrg      1143:
1.10      mrg      1144:        /*
1.106     chs      1145:         * figure out where to put new VM range
1.10      mrg      1146:         */
1.1       mrg      1147:
1.180     yamt     1148: retry:
1.204.2.3  yamt     1149:        if (vm_map_lock_try(map) == false) {
1.204.2.7  yamt     1150:                if ((flags & UVM_FLAG_TRYLOCK) != 0 &&
                   1151:                    (map->flags & VM_MAP_INTRSAFE) == 0) {
1.94      chs      1152:                        return EAGAIN;
1.106     chs      1153:                }
1.10      mrg      1154:                vm_map_lock(map); /* could sleep here */
                   1155:        }
1.204.2.1  yamt     1156:        prev_entry = uvm_map_findspace(map, start, size, &start,
                   1157:            uobj, uoffset, align, flags);
                   1158:        if (prev_entry == NULL) {
1.180     yamt     1159:                unsigned int timestamp;
                   1160:
                   1161:                timestamp = map->timestamp;
                   1162:                UVMHIST_LOG(maphist,"waiting va timestamp=0x%x",
                   1163:                            timestamp,0,0,0);
                   1164:                map->flags |= VM_MAP_WANTVA;
1.10      mrg      1165:                vm_map_unlock(map);
1.180     yamt     1166:
                   1167:                /*
1.204.2.1  yamt     1168:                 * try to reclaim kva and wait until someone does unmap.
1.204.2.4  yamt     1169:                 * fragile locking here, so we awaken every second to
                   1170:                 * recheck the condition.
1.180     yamt     1171:                 */
                   1172:
1.204.2.1  yamt     1173:                vm_map_drain(map, flags);
                   1174:
1.204.2.4  yamt     1175:                mutex_enter(&map->misc_lock);
1.180     yamt     1176:                while ((map->flags & VM_MAP_WANTVA) != 0 &&
                   1177:                   map->timestamp == timestamp) {
1.204.2.1  yamt     1178:                        if ((flags & UVM_FLAG_WAITVA) == 0) {
1.204.2.4  yamt     1179:                                mutex_exit(&map->misc_lock);
1.204.2.1  yamt     1180:                                UVMHIST_LOG(maphist,
                   1181:                                    "<- uvm_map_findspace failed!", 0,0,0,0);
                   1182:                                return ENOMEM;
                   1183:                        } else {
1.204.2.4  yamt     1184:                                cv_timedwait(&map->cv, &map->misc_lock, hz);
1.204.2.1  yamt     1185:                        }
1.180     yamt     1186:                }
1.204.2.4  yamt     1187:                mutex_exit(&map->misc_lock);
1.180     yamt     1188:                goto retry;
1.10      mrg      1189:        }
1.1       mrg      1190:
1.40      thorpej  1191: #ifdef PMAP_GROWKERNEL
1.152     simonb   1192:        /*
                   1193:         * If the kernel pmap can't map the requested space,
                   1194:         * then allocate more resources for it.
                   1195:         */
1.174     yamt     1196:        if (map == kernel_map && uvm_maxkaddr < (start + size))
                   1197:                uvm_maxkaddr = pmap_growkernel(start + size);
1.10      mrg      1198: #endif
                   1199:
1.204.2.1  yamt     1200:        UVMMAP_EVCNT_INCR(map_call);
1.10      mrg      1201:
                   1202:        /*
                   1203:         * if uobj is null, then uoffset is either a VAC hint for PMAP_PREFER
1.98      chs      1204:         * [typically from uvm_map_reserve] or it is UVM_UNKNOWN_OFFSET.   in
                   1205:         * either case we want to zero it  before storing it in the map entry
1.10      mrg      1206:         * (because it looks strange and confusing when debugging...)
1.98      chs      1207:         *
                   1208:         * if uobj is not null
1.10      mrg      1209:         *   if uoffset is not UVM_UNKNOWN_OFFSET then we have a normal mapping
                   1210:         *      and we do not need to change uoffset.
                   1211:         *   if uoffset is UVM_UNKNOWN_OFFSET then we need to find the offset
                   1212:         *      now (based on the starting address of the map).   this case is
                   1213:         *      for kernel object mappings where we don't know the offset until
                   1214:         *      the virtual address is found (with uvm_map_findspace).   the
                   1215:         *      offset is the distance we are from the start of the map.
                   1216:         */
                   1217:
                   1218:        if (uobj == NULL) {
                   1219:                uoffset = 0;
                   1220:        } else {
                   1221:                if (uoffset == UVM_UNKNOWN_OFFSET) {
1.85      chs      1222:                        KASSERT(UVM_OBJ_IS_KERN_OBJECT(uobj));
1.174     yamt     1223:                        uoffset = start - vm_map_min(kernel_map);
1.10      mrg      1224:                }
                   1225:        }
                   1226:
1.174     yamt     1227:        args->uma_flags = flags;
                   1228:        args->uma_prev = prev_entry;
                   1229:        args->uma_start = start;
                   1230:        args->uma_size = size;
                   1231:        args->uma_uobj = uobj;
                   1232:        args->uma_uoffset = uoffset;
                   1233:
                   1234:        return 0;
                   1235: }
                   1236:
                   1237: int
                   1238: uvm_map_enter(struct vm_map *map, const struct uvm_map_args *args,
                   1239:     struct vm_map_entry *new_entry)
                   1240: {
                   1241:        struct vm_map_entry *prev_entry = args->uma_prev;
                   1242:        struct vm_map_entry *dead = NULL;
                   1243:
                   1244:        const uvm_flag_t flags = args->uma_flags;
                   1245:        const vm_prot_t prot = UVM_PROTECTION(flags);
                   1246:        const vm_prot_t maxprot = UVM_MAXPROTECTION(flags);
                   1247:        const vm_inherit_t inherit = UVM_INHERIT(flags);
                   1248:        const int amapwaitflag = (flags & UVM_FLAG_NOWAIT) ?
                   1249:            AMAP_EXTEND_NOWAIT : 0;
                   1250:        const int advice = UVM_ADVICE(flags);
                   1251:        const int meflagval = (flags & UVM_FLAG_QUANTUM) ?
                   1252:            UVM_MAP_QUANTUM : 0;
                   1253:
                   1254:        vaddr_t start = args->uma_start;
                   1255:        vsize_t size = args->uma_size;
                   1256:        struct uvm_object *uobj = args->uma_uobj;
                   1257:        voff_t uoffset = args->uma_uoffset;
                   1258:
                   1259:        const int kmap = (vm_map_pmap(map) == pmap_kernel());
                   1260:        int merged = 0;
                   1261:        int error;
1.176     yamt     1262:        int newetype;
1.174     yamt     1263:
                   1264:        UVMHIST_FUNC("uvm_map_enter");
                   1265:        UVMHIST_CALLED(maphist);
                   1266:
                   1267:        UVMHIST_LOG(maphist, "(map=0x%x, start=0x%x, size=%d, flags=0x%x)",
                   1268:            map, start, size, flags);
                   1269:        UVMHIST_LOG(maphist, "  uobj/offset 0x%x/%d", uobj, uoffset,0,0);
                   1270:
1.204.2.1  yamt     1271:        KASSERT(map->hint == prev_entry); /* bimerge case assumes this */
                   1272:
1.174     yamt     1273:        if (flags & UVM_FLAG_QUANTUM) {
                   1274:                KASSERT(new_entry);
                   1275:                KASSERT(new_entry->flags & UVM_MAP_QUANTUM);
                   1276:        }
                   1277:
1.176     yamt     1278:        if (uobj)
                   1279:                newetype = UVM_ET_OBJ;
                   1280:        else
                   1281:                newetype = 0;
                   1282:
                   1283:        if (flags & UVM_FLAG_COPYONW) {
                   1284:                newetype |= UVM_ET_COPYONWRITE;
                   1285:                if ((flags & UVM_FLAG_OVERLAY) == 0)
                   1286:                        newetype |= UVM_ET_NEEDSCOPY;
                   1287:        }
                   1288:
1.10      mrg      1289:        /*
1.106     chs      1290:         * try and insert in map by extending previous entry, if possible.
1.10      mrg      1291:         * XXX: we don't try and pull back the next entry.   might be useful
                   1292:         * for a stack, but we are currently allocating our stack in advance.
                   1293:         */
                   1294:
1.121     atatat   1295:        if (flags & UVM_FLAG_NOMERGE)
                   1296:                goto nomerge;
                   1297:
1.194     yamt     1298:        if (prev_entry->end == start &&
1.121     atatat   1299:            prev_entry != &map->header &&
1.194     yamt     1300:            UVM_ET_ISCOMPATIBLE(prev_entry, newetype, uobj, meflagval,
                   1301:            prot, maxprot, inherit, advice, 0)) {
1.161     matt     1302:
1.10      mrg      1303:                if (uobj && prev_entry->offset +
                   1304:                    (prev_entry->end - prev_entry->start) != uoffset)
1.121     atatat   1305:                        goto forwardmerge;
1.10      mrg      1306:
                   1307:                /*
1.98      chs      1308:                 * can't extend a shared amap.  note: no need to lock amap to
1.34      chuck    1309:                 * look at refs since we don't care about its exact value.
1.10      mrg      1310:                 * if it is one (i.e. we have only reference) it will stay there
                   1311:                 */
1.85      chs      1312:
1.10      mrg      1313:                if (prev_entry->aref.ar_amap &&
1.34      chuck    1314:                    amap_refs(prev_entry->aref.ar_amap) != 1) {
1.121     atatat   1315:                        goto forwardmerge;
1.10      mrg      1316:                }
1.85      chs      1317:
1.119     chs      1318:                if (prev_entry->aref.ar_amap) {
1.139     enami    1319:                        error = amap_extend(prev_entry, size,
1.126     bouyer   1320:                            amapwaitflag | AMAP_EXTEND_FORWARDS);
1.174     yamt     1321:                        if (error)
1.191     yamt     1322:                                goto nomerge;
1.119     chs      1323:                }
1.10      mrg      1324:
1.121     atatat   1325:                if (kmap)
1.204.2.1  yamt     1326:                        UVMMAP_EVCNT_INCR(kbackmerge);
1.121     atatat   1327:                else
1.204.2.1  yamt     1328:                        UVMMAP_EVCNT_INCR(ubackmerge);
1.10      mrg      1329:                UVMHIST_LOG(maphist,"  starting back merge", 0, 0, 0, 0);
                   1330:
                   1331:                /*
                   1332:                 * drop our reference to uobj since we are extending a reference
                   1333:                 * that we already have (the ref count can not drop to zero).
                   1334:                 */
1.119     chs      1335:
1.10      mrg      1336:                if (uobj && uobj->pgops->pgo_detach)
                   1337:                        uobj->pgops->pgo_detach(uobj);
                   1338:
                   1339:                prev_entry->end += size;
1.145     yamt     1340:                uvm_rb_fixup(map, prev_entry);
                   1341:
1.204.2.1  yamt     1342:                uvm_map_check(map, "map backmerged");
1.10      mrg      1343:
                   1344:                UVMHIST_LOG(maphist,"<- done (via backmerge)!", 0, 0, 0, 0);
1.121     atatat   1345:                merged++;
1.106     chs      1346:        }
1.10      mrg      1347:
1.121     atatat   1348: forwardmerge:
1.194     yamt     1349:        if (prev_entry->next->start == (start + size) &&
1.121     atatat   1350:            prev_entry->next != &map->header &&
1.194     yamt     1351:            UVM_ET_ISCOMPATIBLE(prev_entry->next, newetype, uobj, meflagval,
                   1352:            prot, maxprot, inherit, advice, 0)) {
1.161     matt     1353:
1.121     atatat   1354:                if (uobj && prev_entry->next->offset != uoffset + size)
                   1355:                        goto nomerge;
                   1356:
                   1357:                /*
                   1358:                 * can't extend a shared amap.  note: no need to lock amap to
                   1359:                 * look at refs since we don't care about its exact value.
1.122     atatat   1360:                 * if it is one (i.e. we have only reference) it will stay there.
                   1361:                 *
                   1362:                 * note that we also can't merge two amaps, so if we
                   1363:                 * merged with the previous entry which has an amap,
                   1364:                 * and the next entry also has an amap, we give up.
                   1365:                 *
1.125     atatat   1366:                 * Interesting cases:
                   1367:                 * amap, new, amap -> give up second merge (single fwd extend)
                   1368:                 * amap, new, none -> double forward extend (extend again here)
                   1369:                 * none, new, amap -> double backward extend (done here)
                   1370:                 * uobj, new, amap -> single backward extend (done here)
                   1371:                 *
1.122     atatat   1372:                 * XXX should we attempt to deal with someone refilling
                   1373:                 * the deallocated region between two entries that are
                   1374:                 * backed by the same amap (ie, arefs is 2, "prev" and
                   1375:                 * "next" refer to it, and adding this allocation will
                   1376:                 * close the hole, thus restoring arefs to 1 and
                   1377:                 * deallocating the "next" vm_map_entry)?  -- @@@
1.121     atatat   1378:                 */
                   1379:
                   1380:                if (prev_entry->next->aref.ar_amap &&
1.122     atatat   1381:                    (amap_refs(prev_entry->next->aref.ar_amap) != 1 ||
                   1382:                     (merged && prev_entry->aref.ar_amap))) {
1.121     atatat   1383:                        goto nomerge;
                   1384:                }
                   1385:
1.122     atatat   1386:                if (merged) {
1.123     atatat   1387:                        /*
                   1388:                         * Try to extend the amap of the previous entry to
                   1389:                         * cover the next entry as well.  If it doesn't work
                   1390:                         * just skip on, don't actually give up, since we've
                   1391:                         * already completed the back merge.
                   1392:                         */
1.125     atatat   1393:                        if (prev_entry->aref.ar_amap) {
                   1394:                                if (amap_extend(prev_entry,
                   1395:                                    prev_entry->next->end -
                   1396:                                    prev_entry->next->start,
1.126     bouyer   1397:                                    amapwaitflag | AMAP_EXTEND_FORWARDS))
1.142     enami    1398:                                        goto nomerge;
1.125     atatat   1399:                        }
                   1400:
                   1401:                        /*
                   1402:                         * Try to extend the amap of the *next* entry
                   1403:                         * back to cover the new allocation *and* the
                   1404:                         * previous entry as well (the previous merge
                   1405:                         * didn't have an amap already otherwise we
                   1406:                         * wouldn't be checking here for an amap).  If
                   1407:                         * it doesn't work just skip on, again, don't
                   1408:                         * actually give up, since we've already
                   1409:                         * completed the back merge.
                   1410:                         */
                   1411:                        else if (prev_entry->next->aref.ar_amap) {
                   1412:                                if (amap_extend(prev_entry->next,
                   1413:                                    prev_entry->end -
1.141     atatat   1414:                                    prev_entry->start,
1.126     bouyer   1415:                                    amapwaitflag | AMAP_EXTEND_BACKWARDS))
1.142     enami    1416:                                        goto nomerge;
1.125     atatat   1417:                        }
                   1418:                } else {
                   1419:                        /*
                   1420:                         * Pull the next entry's amap backwards to cover this
                   1421:                         * new allocation.
                   1422:                         */
                   1423:                        if (prev_entry->next->aref.ar_amap) {
                   1424:                                error = amap_extend(prev_entry->next, size,
1.126     bouyer   1425:                                    amapwaitflag | AMAP_EXTEND_BACKWARDS);
1.174     yamt     1426:                                if (error)
1.191     yamt     1427:                                        goto nomerge;
1.125     atatat   1428:                        }
1.122     atatat   1429:                }
                   1430:
1.121     atatat   1431:                if (merged) {
                   1432:                        if (kmap) {
1.204.2.1  yamt     1433:                                UVMMAP_EVCNT_DECR(kbackmerge);
                   1434:                                UVMMAP_EVCNT_INCR(kbimerge);
1.121     atatat   1435:                        } else {
1.204.2.1  yamt     1436:                                UVMMAP_EVCNT_DECR(ubackmerge);
                   1437:                                UVMMAP_EVCNT_INCR(ubimerge);
1.121     atatat   1438:                        }
1.122     atatat   1439:                } else {
1.121     atatat   1440:                        if (kmap)
1.204.2.1  yamt     1441:                                UVMMAP_EVCNT_INCR(kforwmerge);
1.121     atatat   1442:                        else
1.204.2.1  yamt     1443:                                UVMMAP_EVCNT_INCR(uforwmerge);
1.121     atatat   1444:                }
                   1445:                UVMHIST_LOG(maphist,"  starting forward merge", 0, 0, 0, 0);
1.10      mrg      1446:
1.121     atatat   1447:                /*
                   1448:                 * drop our reference to uobj since we are extending a reference
                   1449:                 * that we already have (the ref count can not drop to zero).
                   1450:                 * (if merged, we've already detached)
                   1451:                 */
                   1452:                if (uobj && uobj->pgops->pgo_detach && !merged)
                   1453:                        uobj->pgops->pgo_detach(uobj);
1.1       mrg      1454:
1.121     atatat   1455:                if (merged) {
1.174     yamt     1456:                        dead = prev_entry->next;
1.121     atatat   1457:                        prev_entry->end = dead->end;
                   1458:                        uvm_map_entry_unlink(map, dead);
1.125     atatat   1459:                        if (dead->aref.ar_amap != NULL) {
                   1460:                                prev_entry->aref = dead->aref;
                   1461:                                dead->aref.ar_amap = NULL;
                   1462:                        }
1.121     atatat   1463:                } else {
                   1464:                        prev_entry->next->start -= size;
1.145     yamt     1465:                        if (prev_entry != &map->header)
                   1466:                                uvm_rb_fixup(map, prev_entry);
1.121     atatat   1467:                        if (uobj)
                   1468:                                prev_entry->next->offset = uoffset;
                   1469:                }
1.145     yamt     1470:
1.204.2.1  yamt     1471:                uvm_map_check(map, "map forwardmerged");
1.1       mrg      1472:
1.121     atatat   1473:                UVMHIST_LOG(maphist,"<- done forwardmerge", 0, 0, 0, 0);
                   1474:                merged++;
1.106     chs      1475:        }
1.121     atatat   1476:
                   1477: nomerge:
                   1478:        if (!merged) {
                   1479:                UVMHIST_LOG(maphist,"  allocating new map entry", 0, 0, 0, 0);
                   1480:                if (kmap)
1.204.2.1  yamt     1481:                        UVMMAP_EVCNT_INCR(knomerge);
1.121     atatat   1482:                else
1.204.2.1  yamt     1483:                        UVMMAP_EVCNT_INCR(unomerge);
1.106     chs      1484:
1.10      mrg      1485:                /*
1.121     atatat   1486:                 * allocate new entry and link it in.
1.10      mrg      1487:                 */
1.106     chs      1488:
1.121     atatat   1489:                if (new_entry == NULL) {
1.126     bouyer   1490:                        new_entry = uvm_mapent_alloc(map,
1.127     thorpej  1491:                                (flags & UVM_FLAG_NOWAIT));
1.126     bouyer   1492:                        if (__predict_false(new_entry == NULL)) {
1.174     yamt     1493:                                error = ENOMEM;
                   1494:                                goto done;
1.126     bouyer   1495:                        }
1.121     atatat   1496:                }
1.174     yamt     1497:                new_entry->start = start;
1.121     atatat   1498:                new_entry->end = new_entry->start + size;
                   1499:                new_entry->object.uvm_obj = uobj;
                   1500:                new_entry->offset = uoffset;
                   1501:
1.176     yamt     1502:                new_entry->etype = newetype;
1.121     atatat   1503:
1.161     matt     1504:                if (flags & UVM_FLAG_NOMERGE) {
                   1505:                        new_entry->flags |= UVM_MAP_NOMERGE;
                   1506:                }
1.121     atatat   1507:
                   1508:                new_entry->protection = prot;
                   1509:                new_entry->max_protection = maxprot;
                   1510:                new_entry->inheritance = inherit;
                   1511:                new_entry->wired_count = 0;
                   1512:                new_entry->advice = advice;
                   1513:                if (flags & UVM_FLAG_OVERLAY) {
                   1514:
                   1515:                        /*
                   1516:                         * to_add: for BSS we overallocate a little since we
                   1517:                         * are likely to extend
                   1518:                         */
                   1519:
                   1520:                        vaddr_t to_add = (flags & UVM_FLAG_AMAPPAD) ?
                   1521:                                UVM_AMAP_CHUNK << PAGE_SHIFT : 0;
1.126     bouyer   1522:                        struct vm_amap *amap = amap_alloc(size, to_add,
1.204.2.2  yamt     1523:                            (flags & UVM_FLAG_NOWAIT));
1.126     bouyer   1524:                        if (__predict_false(amap == NULL)) {
1.174     yamt     1525:                                error = ENOMEM;
                   1526:                                goto done;
1.126     bouyer   1527:                        }
1.121     atatat   1528:                        new_entry->aref.ar_pageoff = 0;
                   1529:                        new_entry->aref.ar_amap = amap;
                   1530:                } else {
                   1531:                        new_entry->aref.ar_pageoff = 0;
                   1532:                        new_entry->aref.ar_amap = NULL;
                   1533:                }
                   1534:                uvm_map_entry_link(map, prev_entry, new_entry);
1.1       mrg      1535:
1.121     atatat   1536:                /*
                   1537:                 * Update the free space hint
                   1538:                 */
1.10      mrg      1539:
1.121     atatat   1540:                if ((map->first_free == prev_entry) &&
                   1541:                    (prev_entry->end >= new_entry->start))
                   1542:                        map->first_free = new_entry;
1.174     yamt     1543:
                   1544:                new_entry = NULL;
1.121     atatat   1545:        }
1.10      mrg      1546:
1.146     yamt     1547:        map->size += size;
                   1548:
1.10      mrg      1549:        UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1.174     yamt     1550:
                   1551:        error = 0;
                   1552: done:
1.204.2.7  yamt     1553:        if ((flags & UVM_FLAG_QUANTUM) == 0) {
                   1554:                /*
                   1555:                 * vmk_merged_entries is locked by the map's lock.
                   1556:                 */
                   1557:                vm_map_unlock(map);
                   1558:        }
                   1559:        if (new_entry && error == 0) {
                   1560:                KDASSERT(merged);
                   1561:                uvm_mapent_free_merged(map, new_entry);
                   1562:                new_entry = NULL;
1.174     yamt     1563:        }
                   1564:        if (dead) {
                   1565:                KDASSERT(merged);
1.181     yamt     1566:                uvm_mapent_free_merged(map, dead);
1.174     yamt     1567:        }
1.204.2.7  yamt     1568:        if ((flags & UVM_FLAG_QUANTUM) != 0) {
                   1569:                vm_map_unlock(map);
                   1570:        }
                   1571:        if (new_entry != NULL) {
                   1572:                uvm_mapent_free(new_entry);
                   1573:        }
1.174     yamt     1574:        return error;
1.1       mrg      1575: }
                   1576:
                   1577: /*
1.204.2.7  yamt     1578:  * uvm_map_lookup_entry_bytree: lookup an entry in tree
                   1579:  */
                   1580:
                   1581: static bool
                   1582: uvm_map_lookup_entry_bytree(struct vm_map *map, vaddr_t address,
                   1583:     struct vm_map_entry **entry        /* OUT */)
                   1584: {
                   1585:        struct vm_map_entry *prev = &map->header;
                   1586:        struct vm_map_entry *cur = RB_ROOT(&map->rbhead);
                   1587:
                   1588:        while (cur) {
                   1589:                if (address >= cur->start) {
                   1590:                        if (address < cur->end) {
                   1591:                                *entry = cur;
                   1592:                                return true;
                   1593:                        }
                   1594:                        prev = cur;
                   1595:                        cur = RB_RIGHT(cur, rb_entry);
                   1596:                } else
                   1597:                        cur = RB_LEFT(cur, rb_entry);
                   1598:        }
                   1599:        *entry = prev;
                   1600:        return false;
                   1601: }
                   1602:
                   1603: /*
1.1       mrg      1604:  * uvm_map_lookup_entry: find map entry at or before an address
                   1605:  *
                   1606:  * => map must at least be read-locked by caller
                   1607:  * => entry is returned in "entry"
                   1608:  * => return value is true if address is in the returned entry
                   1609:  */
                   1610:
1.204.2.3  yamt     1611: bool
1.138     enami    1612: uvm_map_lookup_entry(struct vm_map *map, vaddr_t address,
                   1613:     struct vm_map_entry **entry        /* OUT */)
1.1       mrg      1614: {
1.99      chs      1615:        struct vm_map_entry *cur;
1.204.2.3  yamt     1616:        bool use_tree = false;
1.1       mrg      1617:        UVMHIST_FUNC("uvm_map_lookup_entry");
                   1618:        UVMHIST_CALLED(maphist);
                   1619:
                   1620:        UVMHIST_LOG(maphist,"(map=0x%x,addr=0x%x,ent=0x%x)",
1.10      mrg      1621:            map, address, entry, 0);
1.1       mrg      1622:
                   1623:        /*
1.10      mrg      1624:         * start looking either from the head of the
                   1625:         * list, or from the hint.
1.1       mrg      1626:         */
                   1627:
                   1628:        cur = map->hint;
                   1629:
                   1630:        if (cur == &map->header)
                   1631:                cur = cur->next;
                   1632:
1.204.2.1  yamt     1633:        UVMMAP_EVCNT_INCR(mlk_call);
1.1       mrg      1634:        if (address >= cur->start) {
1.99      chs      1635:
1.139     enami    1636:                /*
1.10      mrg      1637:                 * go from hint to end of list.
1.1       mrg      1638:                 *
1.10      mrg      1639:                 * but first, make a quick check to see if
                   1640:                 * we are already looking at the entry we
                   1641:                 * want (which is usually the case).
                   1642:                 * note also that we don't need to save the hint
                   1643:                 * here... it is the same hint (unless we are
                   1644:                 * at the header, in which case the hint didn't
                   1645:                 * buy us anything anyway).
1.1       mrg      1646:                 */
1.99      chs      1647:
1.144     yamt     1648:                if (cur != &map->header && cur->end > address) {
1.204.2.1  yamt     1649:                        UVMMAP_EVCNT_INCR(mlk_hint);
1.1       mrg      1650:                        *entry = cur;
                   1651:                        UVMHIST_LOG(maphist,"<- got it via hint (0x%x)",
1.10      mrg      1652:                            cur, 0, 0, 0);
1.204.2.1  yamt     1653:                        uvm_mapent_check(*entry);
1.204.2.3  yamt     1654:                        return (true);
1.1       mrg      1655:                }
1.144     yamt     1656:
                   1657:                if (map->nentries > 30)
1.204.2.3  yamt     1658:                        use_tree = true;
1.10      mrg      1659:        } else {
1.99      chs      1660:
1.139     enami    1661:                /*
1.144     yamt     1662:                 * invalid hint.  use tree.
1.1       mrg      1663:                 */
1.204.2.3  yamt     1664:                use_tree = true;
1.144     yamt     1665:        }
                   1666:
1.204.2.1  yamt     1667:        uvm_map_check(map, __func__);
1.144     yamt     1668:
                   1669:        if (use_tree) {
                   1670:                /*
                   1671:                 * Simple lookup in the tree.  Happens when the hint is
                   1672:                 * invalid, or nentries reach a threshold.
                   1673:                 */
1.204.2.7  yamt     1674:                if (uvm_map_lookup_entry_bytree(map, address, entry)) {
                   1675:                        goto got;
                   1676:                } else {
                   1677:                        goto failed;
1.144     yamt     1678:                }
1.1       mrg      1679:        }
                   1680:
                   1681:        /*
1.10      mrg      1682:         * search linearly
1.1       mrg      1683:         */
                   1684:
1.144     yamt     1685:        while (cur != &map->header) {
1.1       mrg      1686:                if (cur->end > address) {
                   1687:                        if (address >= cur->start) {
1.139     enami    1688:                                /*
1.10      mrg      1689:                                 * save this lookup for future
                   1690:                                 * hints, and return
1.1       mrg      1691:                                 */
                   1692:
                   1693:                                *entry = cur;
1.144     yamt     1694: got:
                   1695:                                SAVE_HINT(map, map->hint, *entry);
1.1       mrg      1696:                                UVMHIST_LOG(maphist,"<- search got it (0x%x)",
1.10      mrg      1697:                                        cur, 0, 0, 0);
1.144     yamt     1698:                                KDASSERT((*entry)->start <= address);
                   1699:                                KDASSERT(address < (*entry)->end);
1.204.2.1  yamt     1700:                                uvm_mapent_check(*entry);
1.204.2.3  yamt     1701:                                return (true);
1.1       mrg      1702:                        }
                   1703:                        break;
                   1704:                }
                   1705:                cur = cur->next;
                   1706:        }
                   1707:        *entry = cur->prev;
1.144     yamt     1708: failed:
1.82      thorpej  1709:        SAVE_HINT(map, map->hint, *entry);
1.1       mrg      1710:        UVMHIST_LOG(maphist,"<- failed!",0,0,0,0);
1.147     yamt     1711:        KDASSERT((*entry) == &map->header || (*entry)->end <= address);
1.144     yamt     1712:        KDASSERT((*entry)->next == &map->header ||
                   1713:            address < (*entry)->next->start);
1.204.2.3  yamt     1714:        return (false);
1.1       mrg      1715: }
                   1716:
                   1717: /*
1.140     enami    1718:  * See if the range between start and start + length fits in the gap
                   1719:  * entry->next->start and entry->end.  Returns 1 if fits, 0 if doesn't
                   1720:  * fit, and -1 address wraps around.
                   1721:  */
1.203     thorpej  1722: static int
1.140     enami    1723: uvm_map_space_avail(vaddr_t *start, vsize_t length, voff_t uoffset,
                   1724:     vsize_t align, int topdown, struct vm_map_entry *entry)
                   1725: {
                   1726:        vaddr_t end;
                   1727:
                   1728: #ifdef PMAP_PREFER
                   1729:        /*
                   1730:         * push start address forward as needed to avoid VAC alias problems.
                   1731:         * we only do this if a valid offset is specified.
                   1732:         */
                   1733:
                   1734:        if (uoffset != UVM_UNKNOWN_OFFSET)
1.182     atatat   1735:                PMAP_PREFER(uoffset, start, length, topdown);
1.140     enami    1736: #endif
                   1737:        if (align != 0) {
                   1738:                if ((*start & (align - 1)) != 0) {
                   1739:                        if (topdown)
                   1740:                                *start &= ~(align - 1);
                   1741:                        else
                   1742:                                *start = roundup(*start, align);
                   1743:                }
                   1744:                /*
                   1745:                 * XXX Should we PMAP_PREFER() here again?
1.182     atatat   1746:                 * eh...i think we're okay
1.140     enami    1747:                 */
                   1748:        }
                   1749:
                   1750:        /*
                   1751:         * Find the end of the proposed new region.  Be sure we didn't
                   1752:         * wrap around the address; if so, we lose.  Otherwise, if the
                   1753:         * proposed new region fits before the next entry, we win.
                   1754:         */
                   1755:
                   1756:        end = *start + length;
                   1757:        if (end < *start)
                   1758:                return (-1);
                   1759:
                   1760:        if (entry->next->start >= end && *start >= entry->end)
                   1761:                return (1);
                   1762:
                   1763:        return (0);
                   1764: }
                   1765:
                   1766: /*
1.1       mrg      1767:  * uvm_map_findspace: find "length" sized space in "map".
                   1768:  *
1.167     junyoung 1769:  * => "hint" is a hint about where we want it, unless UVM_FLAG_FIXED is
                   1770:  *     set in "flags" (in which case we insist on using "hint").
1.1       mrg      1771:  * => "result" is VA returned
                   1772:  * => uobj/uoffset are to be used to handle VAC alignment, if required
1.167     junyoung 1773:  * => if "align" is non-zero, we attempt to align to that value.
1.1       mrg      1774:  * => caller must at least have read-locked map
                   1775:  * => returns NULL on failure, or pointer to prev. map entry if success
                   1776:  * => note this is a cross between the old vm_map_findspace and vm_map_find
                   1777:  */
                   1778:
1.99      chs      1779: struct vm_map_entry *
1.138     enami    1780: uvm_map_findspace(struct vm_map *map, vaddr_t hint, vsize_t length,
                   1781:     vaddr_t *result /* OUT */, struct uvm_object *uobj, voff_t uoffset,
                   1782:     vsize_t align, int flags)
1.1       mrg      1783: {
1.140     enami    1784:        struct vm_map_entry *entry;
1.144     yamt     1785:        struct vm_map_entry *child, *prev, *tmp;
1.140     enami    1786:        vaddr_t orig_hint;
1.131     atatat   1787:        const int topdown = map->flags & VM_MAP_TOPDOWN;
1.1       mrg      1788:        UVMHIST_FUNC("uvm_map_findspace");
                   1789:        UVMHIST_CALLED(maphist);
                   1790:
1.98      chs      1791:        UVMHIST_LOG(maphist, "(map=0x%x, hint=0x%x, len=%d, flags=0x%x)",
1.140     enami    1792:            map, hint, length, flags);
1.85      chs      1793:        KASSERT((align & (align - 1)) == 0);
                   1794:        KASSERT((flags & UVM_FLAG_FIXED) == 0 || align == 0);
1.81      thorpej  1795:
1.204.2.1  yamt     1796:        uvm_map_check(map, "map_findspace entry");
1.144     yamt     1797:
1.81      thorpej  1798:        /*
                   1799:         * remember the original hint.  if we are aligning, then we
                   1800:         * may have to try again with no alignment constraint if
                   1801:         * we fail the first time.
                   1802:         */
1.85      chs      1803:
1.81      thorpej  1804:        orig_hint = hint;
1.184     chs      1805:        if (hint < vm_map_min(map)) {   /* check ranges ... */
1.81      thorpej  1806:                if (flags & UVM_FLAG_FIXED) {
1.1       mrg      1807:                        UVMHIST_LOG(maphist,"<- VA below map range",0,0,0,0);
1.139     enami    1808:                        return (NULL);
1.1       mrg      1809:                }
1.184     chs      1810:                hint = vm_map_min(map);
1.1       mrg      1811:        }
1.184     chs      1812:        if (hint > vm_map_max(map)) {
1.1       mrg      1813:                UVMHIST_LOG(maphist,"<- VA 0x%x > range [0x%x->0x%x]",
1.184     chs      1814:                    hint, vm_map_min(map), vm_map_max(map), 0);
1.139     enami    1815:                return (NULL);
1.1       mrg      1816:        }
                   1817:
                   1818:        /*
                   1819:         * Look for the first possible address; if there's already
                   1820:         * something at this address, we have to start after it.
                   1821:         */
                   1822:
1.131     atatat   1823:        /*
                   1824:         * @@@: there are four, no, eight cases to consider.
                   1825:         *
                   1826:         * 0: found,     fixed,     bottom up -> fail
                   1827:         * 1: found,     fixed,     top down  -> fail
1.140     enami    1828:         * 2: found,     not fixed, bottom up -> start after entry->end,
                   1829:         *                                       loop up
                   1830:         * 3: found,     not fixed, top down  -> start before entry->start,
                   1831:         *                                       loop down
                   1832:         * 4: not found, fixed,     bottom up -> check entry->next->start, fail
                   1833:         * 5: not found, fixed,     top down  -> check entry->next->start, fail
                   1834:         * 6: not found, not fixed, bottom up -> check entry->next->start,
                   1835:         *                                       loop up
                   1836:         * 7: not found, not fixed, top down  -> check entry->next->start,
                   1837:         *                                       loop down
1.131     atatat   1838:         *
                   1839:         * as you can see, it reduces to roughly five cases, and that
                   1840:         * adding top down mapping only adds one unique case (without
                   1841:         * it, there would be four cases).
                   1842:         */
                   1843:
1.184     chs      1844:        if ((flags & UVM_FLAG_FIXED) == 0 && hint == vm_map_min(map)) {
1.140     enami    1845:                entry = map->first_free;
1.1       mrg      1846:        } else {
1.140     enami    1847:                if (uvm_map_lookup_entry(map, hint, &entry)) {
1.1       mrg      1848:                        /* "hint" address already in use ... */
1.81      thorpej  1849:                        if (flags & UVM_FLAG_FIXED) {
1.140     enami    1850:                                UVMHIST_LOG(maphist, "<- fixed & VA in use",
1.10      mrg      1851:                                    0, 0, 0, 0);
1.139     enami    1852:                                return (NULL);
1.1       mrg      1853:                        }
1.140     enami    1854:                        if (topdown)
                   1855:                                /* Start from lower gap. */
                   1856:                                entry = entry->prev;
                   1857:                } else if (flags & UVM_FLAG_FIXED) {
                   1858:                        if (entry->next->start >= hint + length &&
                   1859:                            hint + length > hint)
                   1860:                                goto found;
                   1861:
                   1862:                        /* "hint" address is gap but too small */
                   1863:                        UVMHIST_LOG(maphist, "<- fixed mapping failed",
                   1864:                            0, 0, 0, 0);
                   1865:                        return (NULL); /* only one shot at it ... */
                   1866:                } else {
                   1867:                        /*
                   1868:                         * See if given hint fits in this gap.
                   1869:                         */
                   1870:                        switch (uvm_map_space_avail(&hint, length,
                   1871:                            uoffset, align, topdown, entry)) {
                   1872:                        case 1:
                   1873:                                goto found;
                   1874:                        case -1:
                   1875:                                goto wraparound;
                   1876:                        }
                   1877:
1.148     yamt     1878:                        if (topdown) {
1.140     enami    1879:                                /*
                   1880:                                 * Still there is a chance to fit
                   1881:                                 * if hint > entry->end.
                   1882:                                 */
1.148     yamt     1883:                        } else {
1.168     junyoung 1884:                                /* Start from higher gap. */
1.148     yamt     1885:                                entry = entry->next;
                   1886:                                if (entry == &map->header)
                   1887:                                        goto notfound;
1.140     enami    1888:                                goto nextgap;
1.148     yamt     1889:                        }
1.1       mrg      1890:                }
                   1891:        }
                   1892:
                   1893:        /*
1.144     yamt     1894:         * Note that all UVM_FLAGS_FIXED case is already handled.
                   1895:         */
                   1896:        KDASSERT((flags & UVM_FLAG_FIXED) == 0);
                   1897:
                   1898:        /* Try to find the space in the red-black tree */
                   1899:
                   1900:        /* Check slot before any entry */
                   1901:        hint = topdown ? entry->next->start - length : entry->end;
                   1902:        switch (uvm_map_space_avail(&hint, length, uoffset, align,
                   1903:            topdown, entry)) {
                   1904:        case 1:
                   1905:                goto found;
                   1906:        case -1:
                   1907:                goto wraparound;
                   1908:        }
                   1909:
                   1910: nextgap:
1.148     yamt     1911:        KDASSERT((flags & UVM_FLAG_FIXED) == 0);
1.144     yamt     1912:        /* If there is not enough space in the whole tree, we fail */
                   1913:        tmp = RB_ROOT(&map->rbhead);
                   1914:        if (tmp == NULL || tmp->space < length)
                   1915:                goto notfound;
                   1916:
                   1917:        prev = NULL; /* previous candidate */
                   1918:
                   1919:        /* Find an entry close to hint that has enough space */
                   1920:        for (; tmp;) {
                   1921:                KASSERT(tmp->next->start == tmp->end + tmp->ownspace);
                   1922:                if (topdown) {
                   1923:                        if (tmp->next->start < hint + length &&
                   1924:                            (prev == NULL || tmp->end > prev->end)) {
                   1925:                                if (tmp->ownspace >= length)
                   1926:                                        prev = tmp;
                   1927:                                else if ((child = RB_LEFT(tmp, rb_entry))
                   1928:                                    != NULL && child->space >= length)
                   1929:                                        prev = tmp;
                   1930:                        }
                   1931:                } else {
                   1932:                        if (tmp->end >= hint &&
                   1933:                            (prev == NULL || tmp->end < prev->end)) {
                   1934:                                if (tmp->ownspace >= length)
                   1935:                                        prev = tmp;
                   1936:                                else if ((child = RB_RIGHT(tmp, rb_entry))
                   1937:                                    != NULL && child->space >= length)
                   1938:                                        prev = tmp;
                   1939:                        }
                   1940:                }
                   1941:                if (tmp->next->start < hint + length)
                   1942:                        child = RB_RIGHT(tmp, rb_entry);
                   1943:                else if (tmp->end > hint)
                   1944:                        child = RB_LEFT(tmp, rb_entry);
                   1945:                else {
                   1946:                        if (tmp->ownspace >= length)
                   1947:                                break;
                   1948:                        if (topdown)
                   1949:                                child = RB_LEFT(tmp, rb_entry);
                   1950:                        else
                   1951:                                child = RB_RIGHT(tmp, rb_entry);
                   1952:                }
                   1953:                if (child == NULL || child->space < length)
                   1954:                        break;
                   1955:                tmp = child;
                   1956:        }
                   1957:
1.148     yamt     1958:        if (tmp != NULL && tmp->start < hint && hint < tmp->next->start) {
1.164     junyoung 1959:                /*
1.144     yamt     1960:                 * Check if the entry that we found satifies the
                   1961:                 * space requirement
                   1962:                 */
1.148     yamt     1963:                if (topdown) {
1.149     yamt     1964:                        if (hint > tmp->next->start - length)
                   1965:                                hint = tmp->next->start - length;
1.148     yamt     1966:                } else {
1.149     yamt     1967:                        if (hint < tmp->end)
                   1968:                                hint = tmp->end;
1.148     yamt     1969:                }
                   1970:                switch (uvm_map_space_avail(&hint, length, uoffset, align,
                   1971:                    topdown, tmp)) {
                   1972:                case 1:
1.144     yamt     1973:                        entry = tmp;
                   1974:                        goto found;
1.148     yamt     1975:                case -1:
                   1976:                        goto wraparound;
1.144     yamt     1977:                }
                   1978:                if (tmp->ownspace >= length)
                   1979:                        goto listsearch;
                   1980:        }
                   1981:        if (prev == NULL)
                   1982:                goto notfound;
                   1983:
1.148     yamt     1984:        if (topdown) {
1.150     yamt     1985:                KASSERT(orig_hint >= prev->next->start - length ||
1.148     yamt     1986:                    prev->next->start - length > prev->next->start);
                   1987:                hint = prev->next->start - length;
                   1988:        } else {
1.150     yamt     1989:                KASSERT(orig_hint <= prev->end);
1.148     yamt     1990:                hint = prev->end;
                   1991:        }
                   1992:        switch (uvm_map_space_avail(&hint, length, uoffset, align,
                   1993:            topdown, prev)) {
                   1994:        case 1:
1.144     yamt     1995:                entry = prev;
                   1996:                goto found;
1.148     yamt     1997:        case -1:
                   1998:                goto wraparound;
1.144     yamt     1999:        }
                   2000:        if (prev->ownspace >= length)
                   2001:                goto listsearch;
1.164     junyoung 2002:
1.144     yamt     2003:        if (topdown)
                   2004:                tmp = RB_LEFT(prev, rb_entry);
                   2005:        else
                   2006:                tmp = RB_RIGHT(prev, rb_entry);
                   2007:        for (;;) {
                   2008:                KASSERT(tmp && tmp->space >= length);
                   2009:                if (topdown)
                   2010:                        child = RB_RIGHT(tmp, rb_entry);
                   2011:                else
                   2012:                        child = RB_LEFT(tmp, rb_entry);
                   2013:                if (child && child->space >= length) {
                   2014:                        tmp = child;
                   2015:                        continue;
                   2016:                }
                   2017:                if (tmp->ownspace >= length)
                   2018:                        break;
                   2019:                if (topdown)
                   2020:                        tmp = RB_LEFT(tmp, rb_entry);
                   2021:                else
                   2022:                        tmp = RB_RIGHT(tmp, rb_entry);
                   2023:        }
1.164     junyoung 2024:
1.148     yamt     2025:        if (topdown) {
1.150     yamt     2026:                KASSERT(orig_hint >= tmp->next->start - length ||
1.148     yamt     2027:                    tmp->next->start - length > tmp->next->start);
                   2028:                hint = tmp->next->start - length;
                   2029:        } else {
1.150     yamt     2030:                KASSERT(orig_hint <= tmp->end);
1.148     yamt     2031:                hint = tmp->end;
                   2032:        }
1.144     yamt     2033:        switch (uvm_map_space_avail(&hint, length, uoffset, align,
                   2034:            topdown, tmp)) {
                   2035:        case 1:
                   2036:                entry = tmp;
                   2037:                goto found;
1.148     yamt     2038:        case -1:
                   2039:                goto wraparound;
1.144     yamt     2040:        }
                   2041:
1.164     junyoung 2042:        /*
1.144     yamt     2043:         * The tree fails to find an entry because of offset or alignment
                   2044:         * restrictions.  Search the list instead.
                   2045:         */
                   2046:  listsearch:
                   2047:        /*
1.1       mrg      2048:         * Look through the rest of the map, trying to fit a new region in
                   2049:         * the gap between existing regions, or after the very last region.
1.140     enami    2050:         * note: entry->end = base VA of current gap,
                   2051:         *       entry->next->start = VA of end of current gap
1.1       mrg      2052:         */
1.99      chs      2053:
1.140     enami    2054:        for (;;) {
                   2055:                /* Update hint for current gap. */
                   2056:                hint = topdown ? entry->next->start - length : entry->end;
                   2057:
                   2058:                /* See if it fits. */
                   2059:                switch (uvm_map_space_avail(&hint, length, uoffset, align,
                   2060:                    topdown, entry)) {
                   2061:                case 1:
                   2062:                        goto found;
                   2063:                case -1:
                   2064:                        goto wraparound;
                   2065:                }
                   2066:
                   2067:                /* Advance to next/previous gap */
                   2068:                if (topdown) {
                   2069:                        if (entry == &map->header) {
                   2070:                                UVMHIST_LOG(maphist, "<- failed (off start)",
                   2071:                                    0,0,0,0);
                   2072:                                goto notfound;
1.134     matt     2073:                        }
1.140     enami    2074:                        entry = entry->prev;
                   2075:                } else {
                   2076:                        entry = entry->next;
                   2077:                        if (entry == &map->header) {
                   2078:                                UVMHIST_LOG(maphist, "<- failed (off end)",
1.81      thorpej  2079:                                    0,0,0,0);
1.140     enami    2080:                                goto notfound;
1.81      thorpej  2081:                        }
1.1       mrg      2082:                }
                   2083:        }
1.140     enami    2084:
                   2085:  found:
1.82      thorpej  2086:        SAVE_HINT(map, map->hint, entry);
1.1       mrg      2087:        *result = hint;
                   2088:        UVMHIST_LOG(maphist,"<- got it!  (result=0x%x)", hint, 0,0,0);
1.148     yamt     2089:        KASSERT( topdown || hint >= orig_hint);
                   2090:        KASSERT(!topdown || hint <= orig_hint);
1.144     yamt     2091:        KASSERT(entry->end <= hint);
                   2092:        KASSERT(hint + length <= entry->next->start);
1.1       mrg      2093:        return (entry);
1.140     enami    2094:
                   2095:  wraparound:
                   2096:        UVMHIST_LOG(maphist, "<- failed (wrap around)", 0,0,0,0);
                   2097:
1.165     yamt     2098:        return (NULL);
                   2099:
1.140     enami    2100:  notfound:
1.165     yamt     2101:        UVMHIST_LOG(maphist, "<- failed (notfound)", 0,0,0,0);
                   2102:
1.140     enami    2103:        return (NULL);
1.1       mrg      2104: }
                   2105:
                   2106: /*
                   2107:  *   U N M A P   -   m a i n   h e l p e r   f u n c t i o n s
                   2108:  */
                   2109:
                   2110: /*
                   2111:  * uvm_unmap_remove: remove mappings from a vm_map (from "start" up to "stop")
                   2112:  *
1.98      chs      2113:  * => caller must check alignment and size
1.1       mrg      2114:  * => map must be locked by caller
                   2115:  * => we return a list of map entries that we've remove from the map
                   2116:  *    in "entry_list"
                   2117:  */
                   2118:
1.94      chs      2119: void
1.138     enami    2120: uvm_unmap_remove(struct vm_map *map, vaddr_t start, vaddr_t end,
1.174     yamt     2121:     struct vm_map_entry **entry_list /* OUT */,
1.187     yamt     2122:     struct uvm_mapent_reservation *umr, int flags)
1.10      mrg      2123: {
1.99      chs      2124:        struct vm_map_entry *entry, *first_entry, *next;
1.24      eeh      2125:        vaddr_t len;
1.99      chs      2126:        UVMHIST_FUNC("uvm_unmap_remove"); UVMHIST_CALLED(maphist);
1.10      mrg      2127:
                   2128:        UVMHIST_LOG(maphist,"(map=0x%x, start=0x%x, end=0x%x)",
                   2129:            map, start, end, 0);
                   2130:        VM_MAP_RANGE_CHECK(map, start, end);
                   2131:
1.204.2.1  yamt     2132:        uvm_map_check(map, "unmap_remove entry");
1.144     yamt     2133:
1.10      mrg      2134:        /*
                   2135:         * find first entry
                   2136:         */
1.99      chs      2137:
1.204.2.3  yamt     2138:        if (uvm_map_lookup_entry(map, start, &first_entry) == true) {
1.29      chuck    2139:                /* clip and go... */
1.10      mrg      2140:                entry = first_entry;
1.174     yamt     2141:                UVM_MAP_CLIP_START(map, entry, start, umr);
1.10      mrg      2142:                /* critical!  prevents stale hint */
1.82      thorpej  2143:                SAVE_HINT(map, entry, entry->prev);
1.10      mrg      2144:        } else {
                   2145:                entry = first_entry->next;
                   2146:        }
                   2147:
                   2148:        /*
                   2149:         * Save the free space hint
                   2150:         */
                   2151:
1.204.2.1  yamt     2152:        if (map->first_free != &map->header && map->first_free->start >= start)
1.10      mrg      2153:                map->first_free = entry->prev;
                   2154:
                   2155:        /*
                   2156:         * note: we now re-use first_entry for a different task.  we remove
                   2157:         * a number of map entries from the map and save them in a linked
                   2158:         * list headed by "first_entry".  once we remove them from the map
                   2159:         * the caller should unlock the map and drop the references to the
                   2160:         * backing objects [c.f. uvm_unmap_detach].  the object is to
1.100     wiz      2161:         * separate unmapping from reference dropping.  why?
1.10      mrg      2162:         *   [1] the map has to be locked for unmapping
                   2163:         *   [2] the map need not be locked for reference dropping
                   2164:         *   [3] dropping references may trigger pager I/O, and if we hit
                   2165:         *       a pager that does synchronous I/O we may have to wait for it.
                   2166:         *   [4] we would like all waiting for I/O to occur with maps unlocked
1.98      chs      2167:         *       so that we don't block other threads.
1.10      mrg      2168:         */
1.99      chs      2169:
1.10      mrg      2170:        first_entry = NULL;
1.106     chs      2171:        *entry_list = NULL;
1.10      mrg      2172:
                   2173:        /*
1.98      chs      2174:         * break up the area into map entry sized regions and unmap.  note
1.10      mrg      2175:         * that all mappings have to be removed before we can even consider
                   2176:         * dropping references to amaps or VM objects (otherwise we could end
                   2177:         * up with a mapping to a page on the free list which would be very bad)
                   2178:         */
                   2179:
                   2180:        while ((entry != &map->header) && (entry->start < end)) {
1.174     yamt     2181:                KASSERT((entry->flags & UVM_MAP_FIRST) == 0);
                   2182:
                   2183:                UVM_MAP_CLIP_END(map, entry, end, umr);
1.10      mrg      2184:                next = entry->next;
                   2185:                len = entry->end - entry->start;
1.81      thorpej  2186:
1.10      mrg      2187:                /*
                   2188:                 * unwire before removing addresses from the pmap; otherwise
                   2189:                 * unwiring will put the entries back into the pmap (XXX).
                   2190:                 */
1.1       mrg      2191:
1.106     chs      2192:                if (VM_MAPENT_ISWIRED(entry)) {
1.10      mrg      2193:                        uvm_map_entry_unwire(map, entry);
1.106     chs      2194:                }
1.187     yamt     2195:                if (flags & UVM_FLAG_VAONLY) {
                   2196:
                   2197:                        /* nothing */
                   2198:
                   2199:                } else if ((map->flags & VM_MAP_PAGEABLE) == 0) {
1.10      mrg      2200:
1.106     chs      2201:                        /*
                   2202:                         * if the map is non-pageable, any pages mapped there
                   2203:                         * must be wired and entered with pmap_kenter_pa(),
                   2204:                         * and we should free any such pages immediately.
                   2205:                         * this is mostly used for kmem_map and mb_map.
                   2206:                         */
1.99      chs      2207:
1.174     yamt     2208:                        if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
                   2209:                                uvm_km_pgremove_intrsafe(entry->start,
                   2210:                                    entry->end);
                   2211:                                pmap_kremove(entry->start, len);
                   2212:                        }
1.106     chs      2213:                } else if (UVM_ET_ISOBJ(entry) &&
                   2214:                           UVM_OBJ_IS_KERN_OBJECT(entry->object.uvm_obj)) {
1.85      chs      2215:                        KASSERT(vm_map_pmap(map) == pmap_kernel());
1.1       mrg      2216:
1.10      mrg      2217:                        /*
                   2218:                         * note: kernel object mappings are currently used in
                   2219:                         * two ways:
                   2220:                         *  [1] "normal" mappings of pages in the kernel object
                   2221:                         *  [2] uvm_km_valloc'd allocations in which we
                   2222:                         *      pmap_enter in some non-kernel-object page
                   2223:                         *      (e.g. vmapbuf).
                   2224:                         *
                   2225:                         * for case [1], we need to remove the mapping from
                   2226:                         * the pmap and then remove the page from the kernel
                   2227:                         * object (because, once pages in a kernel object are
                   2228:                         * unmapped they are no longer needed, unlike, say,
                   2229:                         * a vnode where you might want the data to persist
                   2230:                         * until flushed out of a queue).
                   2231:                         *
                   2232:                         * for case [2], we need to remove the mapping from
                   2233:                         * the pmap.  there shouldn't be any pages at the
                   2234:                         * specified offset in the kernel object [but it
                   2235:                         * doesn't hurt to call uvm_km_pgremove just to be
                   2236:                         * safe?]
                   2237:                         *
1.98      chs      2238:                         * uvm_km_pgremove currently does the following:
                   2239:                         *   for pages in the kernel object in range:
1.43      thorpej  2240:                         *     - drops the swap slot
1.10      mrg      2241:                         *     - uvm_pagefree the page
                   2242:                         */
                   2243:
                   2244:                        /*
1.43      thorpej  2245:                         * remove mappings from pmap and drop the pages
                   2246:                         * from the object.  offsets are always relative
                   2247:                         * to vm_map_min(kernel_map).
1.10      mrg      2248:                         */
1.99      chs      2249:
1.106     chs      2250:                        pmap_remove(pmap_kernel(), entry->start,
                   2251:                            entry->start + len);
1.187     yamt     2252:                        uvm_km_pgremove(entry->start, entry->end);
1.10      mrg      2253:
                   2254:                        /*
                   2255:                         * null out kernel_object reference, we've just
                   2256:                         * dropped it
                   2257:                         */
1.99      chs      2258:
1.10      mrg      2259:                        entry->etype &= ~UVM_ET_OBJ;
1.106     chs      2260:                        entry->object.uvm_obj = NULL;
                   2261:                } else if (UVM_ET_ISOBJ(entry) || entry->aref.ar_amap) {
1.99      chs      2262:
1.29      chuck    2263:                        /*
1.139     enami    2264:                         * remove mappings the standard way.
                   2265:                         */
1.99      chs      2266:
1.29      chuck    2267:                        pmap_remove(map->pmap, entry->start, entry->end);
1.10      mrg      2268:                }
                   2269:
1.177     yamt     2270: #if defined(DEBUG)
                   2271:                if ((entry->flags & UVM_MAP_KMAPENT) == 0) {
                   2272:
                   2273:                        /*
                   2274:                         * check if there's remaining mapping,
                   2275:                         * which is a bug in caller.
                   2276:                         */
                   2277:
                   2278:                        vaddr_t va;
                   2279:                        for (va = entry->start; va < entry->end;
                   2280:                            va += PAGE_SIZE) {
                   2281:                                if (pmap_extract(vm_map_pmap(map), va, NULL)) {
                   2282:                                        panic("uvm_unmap_remove: has mapping");
                   2283:                                }
                   2284:                        }
1.187     yamt     2285:
                   2286:                        if (VM_MAP_IS_KERNEL(map)) {
                   2287:                                uvm_km_check_empty(entry->start, entry->end,
                   2288:                                    (map->flags & VM_MAP_INTRSAFE) != 0);
                   2289:                        }
1.177     yamt     2290:                }
                   2291: #endif /* defined(DEBUG) */
                   2292:
1.10      mrg      2293:                /*
1.98      chs      2294:                 * remove entry from map and put it on our list of entries
1.106     chs      2295:                 * that we've nuked.  then go to next entry.
1.10      mrg      2296:                 */
1.99      chs      2297:
1.10      mrg      2298:                UVMHIST_LOG(maphist, "  removed map entry 0x%x", entry, 0, 0,0);
1.82      thorpej  2299:
                   2300:                /* critical!  prevents stale hint */
                   2301:                SAVE_HINT(map, entry, entry->prev);
                   2302:
1.10      mrg      2303:                uvm_map_entry_unlink(map, entry);
1.146     yamt     2304:                KASSERT(map->size >= len);
1.10      mrg      2305:                map->size -= len;
1.131     atatat   2306:                entry->prev = NULL;
1.10      mrg      2307:                entry->next = first_entry;
                   2308:                first_entry = entry;
1.106     chs      2309:                entry = next;
1.10      mrg      2310:        }
1.120     chs      2311:        if ((map->flags & VM_MAP_DYING) == 0) {
                   2312:                pmap_update(vm_map_pmap(map));
                   2313:        }
1.10      mrg      2314:
1.204.2.1  yamt     2315:        uvm_map_check(map, "unmap_remove leave");
1.144     yamt     2316:
1.10      mrg      2317:        /*
                   2318:         * now we've cleaned up the map and are ready for the caller to drop
1.98      chs      2319:         * references to the mapped objects.
1.10      mrg      2320:         */
                   2321:
                   2322:        *entry_list = first_entry;
                   2323:        UVMHIST_LOG(maphist,"<- done!", 0, 0, 0, 0);
1.180     yamt     2324:
                   2325:        if (map->flags & VM_MAP_WANTVA) {
1.204.2.4  yamt     2326:                mutex_enter(&map->misc_lock);
1.180     yamt     2327:                map->flags &= ~VM_MAP_WANTVA;
1.204.2.4  yamt     2328:                cv_broadcast(&map->cv);
                   2329:                mutex_exit(&map->misc_lock);
1.180     yamt     2330:        }
1.1       mrg      2331: }
                   2332:
                   2333: /*
                   2334:  * uvm_unmap_detach: drop references in a chain of map entries
                   2335:  *
                   2336:  * => we will free the map entries as we traverse the list.
                   2337:  */
                   2338:
1.10      mrg      2339: void
1.138     enami    2340: uvm_unmap_detach(struct vm_map_entry *first_entry, int flags)
1.1       mrg      2341: {
1.99      chs      2342:        struct vm_map_entry *next_entry;
1.10      mrg      2343:        UVMHIST_FUNC("uvm_unmap_detach"); UVMHIST_CALLED(maphist);
1.1       mrg      2344:
1.10      mrg      2345:        while (first_entry) {
1.85      chs      2346:                KASSERT(!VM_MAPENT_ISWIRED(first_entry));
1.10      mrg      2347:                UVMHIST_LOG(maphist,
1.98      chs      2348:                    "  detach 0x%x: amap=0x%x, obj=0x%x, submap?=%d",
                   2349:                    first_entry, first_entry->aref.ar_amap,
1.29      chuck    2350:                    first_entry->object.uvm_obj,
                   2351:                    UVM_ET_ISSUBMAP(first_entry));
1.1       mrg      2352:
1.10      mrg      2353:                /*
                   2354:                 * drop reference to amap, if we've got one
                   2355:                 */
                   2356:
                   2357:                if (first_entry->aref.ar_amap)
1.85      chs      2358:                        uvm_map_unreference_amap(first_entry, flags);
1.10      mrg      2359:
                   2360:                /*
                   2361:                 * drop reference to our backing object, if we've got one
                   2362:                 */
1.85      chs      2363:
1.120     chs      2364:                KASSERT(!UVM_ET_ISSUBMAP(first_entry));
                   2365:                if (UVM_ET_ISOBJ(first_entry) &&
                   2366:                    first_entry->object.uvm_obj->pgops->pgo_detach) {
                   2367:                        (*first_entry->object.uvm_obj->pgops->pgo_detach)
                   2368:                                (first_entry->object.uvm_obj);
1.10      mrg      2369:                }
                   2370:                next_entry = first_entry->next;
                   2371:                uvm_mapent_free(first_entry);
                   2372:                first_entry = next_entry;
                   2373:        }
                   2374:        UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
1.1       mrg      2375: }
                   2376:
                   2377: /*
                   2378:  *   E X T R A C T I O N   F U N C T I O N S
                   2379:  */
                   2380:
1.98      chs      2381: /*
1.1       mrg      2382:  * uvm_map_reserve: reserve space in a vm_map for future use.
                   2383:  *
1.98      chs      2384:  * => we reserve space in a map by putting a dummy map entry in the
1.1       mrg      2385:  *    map (dummy means obj=NULL, amap=NULL, prot=VM_PROT_NONE)
                   2386:  * => map should be unlocked (we will write lock it)
                   2387:  * => we return true if we were able to reserve space
                   2388:  * => XXXCDC: should be inline?
                   2389:  */
                   2390:
1.10      mrg      2391: int
1.138     enami    2392: uvm_map_reserve(struct vm_map *map, vsize_t size,
                   2393:     vaddr_t offset     /* hint for pmap_prefer */,
1.204.2.5  yamt     2394:     vsize_t align      /* alignment */,
1.204.2.1  yamt     2395:     vaddr_t *raddr     /* IN:hint, OUT: reserved VA */,
                   2396:     uvm_flag_t flags   /* UVM_FLAG_FIXED or 0 */)
1.1       mrg      2397: {
1.98      chs      2398:        UVMHIST_FUNC("uvm_map_reserve"); UVMHIST_CALLED(maphist);
1.85      chs      2399:
1.10      mrg      2400:        UVMHIST_LOG(maphist, "(map=0x%x, size=0x%x, offset=0x%x,addr=0x%x)",
1.139     enami    2401:            map,size,offset,raddr);
1.85      chs      2402:
1.10      mrg      2403:        size = round_page(size);
1.85      chs      2404:
1.10      mrg      2405:        /*
                   2406:         * reserve some virtual space.
                   2407:         */
1.85      chs      2408:
1.204.2.5  yamt     2409:        if (uvm_map(map, raddr, size, NULL, offset, align,
1.10      mrg      2410:            UVM_MAPFLAG(UVM_PROT_NONE, UVM_PROT_NONE, UVM_INH_NONE,
1.204.2.1  yamt     2411:            UVM_ADV_RANDOM, UVM_FLAG_NOMERGE|flags)) != 0) {
1.10      mrg      2412:            UVMHIST_LOG(maphist, "<- done (no VM)", 0,0,0,0);
1.204.2.3  yamt     2413:                return (false);
1.98      chs      2414:        }
1.85      chs      2415:
1.10      mrg      2416:        UVMHIST_LOG(maphist, "<- done (*raddr=0x%x)", *raddr,0,0,0);
1.204.2.3  yamt     2417:        return (true);
1.1       mrg      2418: }
                   2419:
                   2420: /*
1.98      chs      2421:  * uvm_map_replace: replace a reserved (blank) area of memory with
1.1       mrg      2422:  * real mappings.
                   2423:  *
1.98      chs      2424:  * => caller must WRITE-LOCK the map
1.204.2.3  yamt     2425:  * => we return true if replacement was a success
1.1       mrg      2426:  * => we expect the newents chain to have nnewents entrys on it and
                   2427:  *    we expect newents->prev to point to the last entry on the list
                   2428:  * => note newents is allowed to be NULL
                   2429:  */
                   2430:
1.10      mrg      2431: int
1.138     enami    2432: uvm_map_replace(struct vm_map *map, vaddr_t start, vaddr_t end,
                   2433:     struct vm_map_entry *newents, int nnewents)
1.10      mrg      2434: {
1.99      chs      2435:        struct vm_map_entry *oldent, *last;
1.1       mrg      2436:
1.204.2.1  yamt     2437:        uvm_map_check(map, "map_replace entry");
1.144     yamt     2438:
1.10      mrg      2439:        /*
                   2440:         * first find the blank map entry at the specified address
                   2441:         */
1.85      chs      2442:
1.10      mrg      2443:        if (!uvm_map_lookup_entry(map, start, &oldent)) {
1.204.2.3  yamt     2444:                return (false);
1.10      mrg      2445:        }
1.85      chs      2446:
1.10      mrg      2447:        /*
                   2448:         * check to make sure we have a proper blank entry
                   2449:         */
1.1       mrg      2450:
1.204.2.1  yamt     2451:        if (end < oldent->end && !VM_MAP_USE_KMAPENT(map)) {
                   2452:                UVM_MAP_CLIP_END(map, oldent, end, NULL);
                   2453:        }
1.98      chs      2454:        if (oldent->start != start || oldent->end != end ||
1.10      mrg      2455:            oldent->object.uvm_obj != NULL || oldent->aref.ar_amap != NULL) {
1.204.2.3  yamt     2456:                return (false);
1.10      mrg      2457:        }
1.1       mrg      2458:
                   2459: #ifdef DIAGNOSTIC
1.99      chs      2460:
1.10      mrg      2461:        /*
                   2462:         * sanity check the newents chain
                   2463:         */
1.99      chs      2464:
1.10      mrg      2465:        {
1.99      chs      2466:                struct vm_map_entry *tmpent = newents;
1.10      mrg      2467:                int nent = 0;
1.24      eeh      2468:                vaddr_t cur = start;
1.10      mrg      2469:
                   2470:                while (tmpent) {
                   2471:                        nent++;
                   2472:                        if (tmpent->start < cur)
                   2473:                                panic("uvm_map_replace1");
                   2474:                        if (tmpent->start > tmpent->end || tmpent->end > end) {
                   2475:                printf("tmpent->start=0x%lx, tmpent->end=0x%lx, end=0x%lx\n",
                   2476:                            tmpent->start, tmpent->end, end);
                   2477:                                panic("uvm_map_replace2");
                   2478:                        }
                   2479:                        cur = tmpent->end;
                   2480:                        if (tmpent->next) {
                   2481:                                if (tmpent->next->prev != tmpent)
                   2482:                                        panic("uvm_map_replace3");
                   2483:                        } else {
                   2484:                                if (newents->prev != tmpent)
                   2485:                                        panic("uvm_map_replace4");
                   2486:                        }
                   2487:                        tmpent = tmpent->next;
                   2488:                }
                   2489:                if (nent != nnewents)
                   2490:                        panic("uvm_map_replace5");
                   2491:        }
                   2492: #endif
                   2493:
                   2494:        /*
                   2495:         * map entry is a valid blank!   replace it.   (this does all the
                   2496:         * work of map entry link/unlink...).
                   2497:         */
                   2498:
                   2499:        if (newents) {
1.99      chs      2500:                last = newents->prev;
1.10      mrg      2501:
                   2502:                /* critical: flush stale hints out of map */
1.82      thorpej  2503:                SAVE_HINT(map, map->hint, newents);
1.10      mrg      2504:                if (map->first_free == oldent)
                   2505:                        map->first_free = last;
                   2506:
                   2507:                last->next = oldent->next;
                   2508:                last->next->prev = last;
1.144     yamt     2509:
                   2510:                /* Fix RB tree */
                   2511:                uvm_rb_remove(map, oldent);
                   2512:
1.10      mrg      2513:                newents->prev = oldent->prev;
                   2514:                newents->prev->next = newents;
                   2515:                map->nentries = map->nentries + (nnewents - 1);
                   2516:
1.144     yamt     2517:                /* Fixup the RB tree */
                   2518:                {
                   2519:                        int i;
                   2520:                        struct vm_map_entry *tmp;
                   2521:
                   2522:                        tmp = newents;
                   2523:                        for (i = 0; i < nnewents && tmp; i++) {
                   2524:                                uvm_rb_insert(map, tmp);
                   2525:                                tmp = tmp->next;
                   2526:                        }
                   2527:                }
1.10      mrg      2528:        } else {
                   2529:                /* NULL list of new entries: just remove the old one */
1.204.2.1  yamt     2530:                clear_hints(map, oldent);
1.10      mrg      2531:                uvm_map_entry_unlink(map, oldent);
                   2532:        }
                   2533:
1.204.2.1  yamt     2534:        uvm_map_check(map, "map_replace leave");
1.10      mrg      2535:
                   2536:        /*
1.204.2.1  yamt     2537:         * now we can free the old blank entry and return.
1.10      mrg      2538:         */
1.1       mrg      2539:
1.10      mrg      2540:        uvm_mapent_free(oldent);
1.204.2.3  yamt     2541:        return (true);
1.1       mrg      2542: }
                   2543:
                   2544: /*
                   2545:  * uvm_map_extract: extract a mapping from a map and put it somewhere
                   2546:  *     (maybe removing the old mapping)
                   2547:  *
                   2548:  * => maps should be unlocked (we will write lock them)
                   2549:  * => returns 0 on success, error code otherwise
                   2550:  * => start must be page aligned
                   2551:  * => len must be page sized
                   2552:  * => flags:
                   2553:  *      UVM_EXTRACT_REMOVE: remove mappings from srcmap
                   2554:  *      UVM_EXTRACT_CONTIG: abort if unmapped area (advisory only)
                   2555:  *      UVM_EXTRACT_QREF: for a temporary extraction do quick obj refs
                   2556:  *      UVM_EXTRACT_FIXPROT: set prot to maxprot as we go
                   2557:  *    >>>NOTE: if you set REMOVE, you are not allowed to use CONTIG or QREF!<<<
                   2558:  *    >>>NOTE: QREF's must be unmapped via the QREF path, thus should only
                   2559:  *             be used from within the kernel in a kernel level map <<<
                   2560:  */
                   2561:
1.10      mrg      2562: int
1.138     enami    2563: uvm_map_extract(struct vm_map *srcmap, vaddr_t start, vsize_t len,
                   2564:     struct vm_map *dstmap, vaddr_t *dstaddrp, int flags)
1.10      mrg      2565: {
1.163     mycroft  2566:        vaddr_t dstaddr, end, newend, oldoffset, fudge, orig_fudge;
1.99      chs      2567:        struct vm_map_entry *chain, *endchain, *entry, *orig_entry, *newentry,
                   2568:            *deadentry, *oldentry;
1.24      eeh      2569:        vsize_t elen;
1.10      mrg      2570:        int nchain, error, copy_ok;
                   2571:        UVMHIST_FUNC("uvm_map_extract"); UVMHIST_CALLED(maphist);
1.85      chs      2572:
1.10      mrg      2573:        UVMHIST_LOG(maphist,"(srcmap=0x%x,start=0x%x, len=0x%x", srcmap, start,
                   2574:            len,0);
                   2575:        UVMHIST_LOG(maphist," ...,dstmap=0x%x, flags=0x%x)", dstmap,flags,0,0);
                   2576:
1.204.2.1  yamt     2577:        uvm_map_check(srcmap, "map_extract src enter");
                   2578:        uvm_map_check(dstmap, "map_extract dst enter");
1.144     yamt     2579:
1.10      mrg      2580:        /*
                   2581:         * step 0: sanity check: start must be on a page boundary, length
                   2582:         * must be page sized.  can't ask for CONTIG/QREF if you asked for
                   2583:         * REMOVE.
                   2584:         */
                   2585:
1.85      chs      2586:        KASSERT((start & PAGE_MASK) == 0 && (len & PAGE_MASK) == 0);
                   2587:        KASSERT((flags & UVM_EXTRACT_REMOVE) == 0 ||
                   2588:                (flags & (UVM_EXTRACT_CONTIG|UVM_EXTRACT_QREF)) == 0);
1.10      mrg      2589:
                   2590:        /*
                   2591:         * step 1: reserve space in the target map for the extracted area
                   2592:         */
                   2593:
1.204.2.1  yamt     2594:        if ((flags & UVM_EXTRACT_RESERVED) == 0) {
                   2595:                dstaddr = vm_map_min(dstmap);
                   2596:                if (!uvm_map_reserve(dstmap, len, start, 0, &dstaddr, 0))
                   2597:                        return (ENOMEM);
                   2598:                *dstaddrp = dstaddr;    /* pass address back to caller */
                   2599:                UVMHIST_LOG(maphist, "  dstaddr=0x%x", dstaddr,0,0,0);
                   2600:        } else {
                   2601:                dstaddr = *dstaddrp;
                   2602:        }
1.10      mrg      2603:
                   2604:        /*
1.98      chs      2605:         * step 2: setup for the extraction process loop by init'ing the
1.10      mrg      2606:         * map entry chain, locking src map, and looking up the first useful
                   2607:         * entry in the map.
                   2608:         */
1.1       mrg      2609:
1.10      mrg      2610:        end = start + len;
                   2611:        newend = dstaddr + len;
                   2612:        chain = endchain = NULL;
                   2613:        nchain = 0;
                   2614:        vm_map_lock(srcmap);
                   2615:
                   2616:        if (uvm_map_lookup_entry(srcmap, start, &entry)) {
                   2617:
                   2618:                /* "start" is within an entry */
                   2619:                if (flags & UVM_EXTRACT_QREF) {
1.85      chs      2620:
1.10      mrg      2621:                        /*
                   2622:                         * for quick references we don't clip the entry, so
                   2623:                         * the entry may map space "before" the starting
                   2624:                         * virtual address... this is the "fudge" factor
                   2625:                         * (which can be non-zero only the first time
                   2626:                         * through the "while" loop in step 3).
                   2627:                         */
1.85      chs      2628:
1.10      mrg      2629:                        fudge = start - entry->start;
                   2630:                } else {
1.85      chs      2631:
1.10      mrg      2632:                        /*
                   2633:                         * normal reference: we clip the map to fit (thus
                   2634:                         * fudge is zero)
                   2635:                         */
1.85      chs      2636:
1.174     yamt     2637:                        UVM_MAP_CLIP_START(srcmap, entry, start, NULL);
1.82      thorpej  2638:                        SAVE_HINT(srcmap, srcmap->hint, entry->prev);
1.10      mrg      2639:                        fudge = 0;
                   2640:                }
1.85      chs      2641:        } else {
1.1       mrg      2642:
1.10      mrg      2643:                /* "start" is not within an entry ... skip to next entry */
                   2644:                if (flags & UVM_EXTRACT_CONTIG) {
                   2645:                        error = EINVAL;
                   2646:                        goto bad;    /* definite hole here ... */
                   2647:                }
1.1       mrg      2648:
1.10      mrg      2649:                entry = entry->next;
                   2650:                fudge = 0;
                   2651:        }
1.85      chs      2652:
1.10      mrg      2653:        /* save values from srcmap for step 6 */
                   2654:        orig_entry = entry;
                   2655:        orig_fudge = fudge;
1.1       mrg      2656:
1.10      mrg      2657:        /*
                   2658:         * step 3: now start looping through the map entries, extracting
                   2659:         * as we go.
                   2660:         */
1.1       mrg      2661:
1.10      mrg      2662:        while (entry->start < end && entry != &srcmap->header) {
1.85      chs      2663:
1.10      mrg      2664:                /* if we are not doing a quick reference, clip it */
                   2665:                if ((flags & UVM_EXTRACT_QREF) == 0)
1.174     yamt     2666:                        UVM_MAP_CLIP_END(srcmap, entry, end, NULL);
1.10      mrg      2667:
                   2668:                /* clear needs_copy (allow chunking) */
                   2669:                if (UVM_ET_ISNEEDSCOPY(entry)) {
1.204.2.1  yamt     2670:                        amap_copy(srcmap, entry,
                   2671:                            AMAP_COPY_NOWAIT|AMAP_COPY_NOMERGE, start, end);
1.10      mrg      2672:                        if (UVM_ET_ISNEEDSCOPY(entry)) {  /* failed? */
                   2673:                                error = ENOMEM;
                   2674:                                goto bad;
                   2675:                        }
1.85      chs      2676:
1.10      mrg      2677:                        /* amap_copy could clip (during chunk)!  update fudge */
                   2678:                        if (fudge) {
1.163     mycroft  2679:                                fudge = start - entry->start;
1.10      mrg      2680:                                orig_fudge = fudge;
                   2681:                        }
                   2682:                }
1.1       mrg      2683:
1.10      mrg      2684:                /* calculate the offset of this from "start" */
                   2685:                oldoffset = (entry->start + fudge) - start;
1.1       mrg      2686:
1.10      mrg      2687:                /* allocate a new map entry */
1.126     bouyer   2688:                newentry = uvm_mapent_alloc(dstmap, 0);
1.10      mrg      2689:                if (newentry == NULL) {
                   2690:                        error = ENOMEM;
                   2691:                        goto bad;
                   2692:                }
                   2693:
                   2694:                /* set up new map entry */
                   2695:                newentry->next = NULL;
                   2696:                newentry->prev = endchain;
                   2697:                newentry->start = dstaddr + oldoffset;
                   2698:                newentry->end =
                   2699:                    newentry->start + (entry->end - (entry->start + fudge));
1.37      chs      2700:                if (newentry->end > newend || newentry->end < newentry->start)
1.10      mrg      2701:                        newentry->end = newend;
                   2702:                newentry->object.uvm_obj = entry->object.uvm_obj;
                   2703:                if (newentry->object.uvm_obj) {
                   2704:                        if (newentry->object.uvm_obj->pgops->pgo_reference)
                   2705:                                newentry->object.uvm_obj->pgops->
                   2706:                                    pgo_reference(newentry->object.uvm_obj);
                   2707:                                newentry->offset = entry->offset + fudge;
                   2708:                } else {
                   2709:                        newentry->offset = 0;
                   2710:                }
                   2711:                newentry->etype = entry->etype;
1.98      chs      2712:                newentry->protection = (flags & UVM_EXTRACT_FIXPROT) ?
                   2713:                        entry->max_protection : entry->protection;
1.10      mrg      2714:                newentry->max_protection = entry->max_protection;
                   2715:                newentry->inheritance = entry->inheritance;
                   2716:                newentry->wired_count = 0;
                   2717:                newentry->aref.ar_amap = entry->aref.ar_amap;
                   2718:                if (newentry->aref.ar_amap) {
1.34      chuck    2719:                        newentry->aref.ar_pageoff =
                   2720:                            entry->aref.ar_pageoff + (fudge >> PAGE_SHIFT);
1.85      chs      2721:                        uvm_map_reference_amap(newentry, AMAP_SHARED |
1.10      mrg      2722:                            ((flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0));
                   2723:                } else {
1.34      chuck    2724:                        newentry->aref.ar_pageoff = 0;
1.10      mrg      2725:                }
                   2726:                newentry->advice = entry->advice;
1.204.2.6  yamt     2727:                if ((flags & UVM_EXTRACT_QREF) != 0) {
                   2728:                        newentry->flags |= UVM_MAP_NOMERGE;
                   2729:                }
1.10      mrg      2730:
                   2731:                /* now link it on the chain */
                   2732:                nchain++;
                   2733:                if (endchain == NULL) {
                   2734:                        chain = endchain = newentry;
                   2735:                } else {
                   2736:                        endchain->next = newentry;
                   2737:                        endchain = newentry;
                   2738:                }
                   2739:
                   2740:                /* end of 'while' loop! */
1.98      chs      2741:                if ((flags & UVM_EXTRACT_CONTIG) && entry->end < end &&
1.10      mrg      2742:                    (entry->next == &srcmap->header ||
                   2743:                    entry->next->start != entry->end)) {
                   2744:                        error = EINVAL;
                   2745:                        goto bad;
                   2746:                }
                   2747:                entry = entry->next;
                   2748:                fudge = 0;
                   2749:        }
                   2750:
                   2751:        /*
                   2752:         * step 4: close off chain (in format expected by uvm_map_replace)
                   2753:         */
                   2754:
                   2755:        if (chain)
                   2756:                chain->prev = endchain;
                   2757:
                   2758:        /*
                   2759:         * step 5: attempt to lock the dest map so we can pmap_copy.
1.98      chs      2760:         * note usage of copy_ok:
1.10      mrg      2761:         *   1 => dstmap locked, pmap_copy ok, and we "replace" here (step 5)
                   2762:         *   0 => dstmap unlocked, NO pmap_copy, and we will "replace" in step 7
                   2763:         */
1.85      chs      2764:
1.204.2.3  yamt     2765:        if (srcmap == dstmap || vm_map_lock_try(dstmap) == true) {
1.10      mrg      2766:                copy_ok = 1;
                   2767:                if (!uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
                   2768:                    nchain)) {
                   2769:                        if (srcmap != dstmap)
                   2770:                                vm_map_unlock(dstmap);
                   2771:                        error = EIO;
                   2772:                        goto bad;
                   2773:                }
                   2774:        } else {
                   2775:                copy_ok = 0;
                   2776:                /* replace defered until step 7 */
                   2777:        }
                   2778:
                   2779:        /*
                   2780:         * step 6: traverse the srcmap a second time to do the following:
                   2781:         *  - if we got a lock on the dstmap do pmap_copy
                   2782:         *  - if UVM_EXTRACT_REMOVE remove the entries
                   2783:         * we make use of orig_entry and orig_fudge (saved in step 2)
                   2784:         */
                   2785:
                   2786:        if (copy_ok || (flags & UVM_EXTRACT_REMOVE)) {
                   2787:
                   2788:                /* purge possible stale hints from srcmap */
                   2789:                if (flags & UVM_EXTRACT_REMOVE) {
1.82      thorpej  2790:                        SAVE_HINT(srcmap, srcmap->hint, orig_entry->prev);
1.204.2.1  yamt     2791:                        if (srcmap->first_free != &srcmap->header &&
                   2792:                            srcmap->first_free->start >= start)
1.10      mrg      2793:                                srcmap->first_free = orig_entry->prev;
                   2794:                }
                   2795:
                   2796:                entry = orig_entry;
                   2797:                fudge = orig_fudge;
                   2798:                deadentry = NULL;       /* for UVM_EXTRACT_REMOVE */
                   2799:
                   2800:                while (entry->start < end && entry != &srcmap->header) {
                   2801:                        if (copy_ok) {
1.74      thorpej  2802:                                oldoffset = (entry->start + fudge) - start;
1.90      chs      2803:                                elen = MIN(end, entry->end) -
1.74      thorpej  2804:                                    (entry->start + fudge);
                   2805:                                pmap_copy(dstmap->pmap, srcmap->pmap,
                   2806:                                    dstaddr + oldoffset, elen,
                   2807:                                    entry->start + fudge);
1.10      mrg      2808:                        }
                   2809:
1.74      thorpej  2810:                        /* we advance "entry" in the following if statement */
1.10      mrg      2811:                        if (flags & UVM_EXTRACT_REMOVE) {
1.98      chs      2812:                                pmap_remove(srcmap->pmap, entry->start,
1.20      chuck    2813:                                                entry->end);
1.139     enami    2814:                                oldentry = entry;       /* save entry */
                   2815:                                entry = entry->next;    /* advance */
1.20      chuck    2816:                                uvm_map_entry_unlink(srcmap, oldentry);
                   2817:                                                        /* add to dead list */
                   2818:                                oldentry->next = deadentry;
                   2819:                                deadentry = oldentry;
1.139     enami    2820:                        } else {
                   2821:                                entry = entry->next;            /* advance */
1.10      mrg      2822:                        }
                   2823:
                   2824:                        /* end of 'while' loop */
                   2825:                        fudge = 0;
                   2826:                }
1.105     chris    2827:                pmap_update(srcmap->pmap);
1.10      mrg      2828:
                   2829:                /*
                   2830:                 * unlock dstmap.  we will dispose of deadentry in
                   2831:                 * step 7 if needed
                   2832:                 */
1.85      chs      2833:
1.10      mrg      2834:                if (copy_ok && srcmap != dstmap)
                   2835:                        vm_map_unlock(dstmap);
                   2836:
1.99      chs      2837:        } else {
                   2838:                deadentry = NULL;
1.10      mrg      2839:        }
                   2840:
                   2841:        /*
                   2842:         * step 7: we are done with the source map, unlock.   if copy_ok
                   2843:         * is 0 then we have not replaced the dummy mapping in dstmap yet
                   2844:         * and we need to do so now.
                   2845:         */
                   2846:
                   2847:        vm_map_unlock(srcmap);
                   2848:        if ((flags & UVM_EXTRACT_REMOVE) && deadentry)
                   2849:                uvm_unmap_detach(deadentry, 0);   /* dispose of old entries */
                   2850:
                   2851:        /* now do the replacement if we didn't do it in step 5 */
                   2852:        if (copy_ok == 0) {
                   2853:                vm_map_lock(dstmap);
                   2854:                error = uvm_map_replace(dstmap, dstaddr, dstaddr+len, chain,
                   2855:                    nchain);
                   2856:                vm_map_unlock(dstmap);
                   2857:
1.204.2.3  yamt     2858:                if (error == false) {
1.10      mrg      2859:                        error = EIO;
                   2860:                        goto bad2;
                   2861:                }
                   2862:        }
1.144     yamt     2863:
1.204.2.1  yamt     2864:        uvm_map_check(srcmap, "map_extract src leave");
                   2865:        uvm_map_check(dstmap, "map_extract dst leave");
1.144     yamt     2866:
1.139     enami    2867:        return (0);
1.10      mrg      2868:
                   2869:        /*
                   2870:         * bad: failure recovery
                   2871:         */
                   2872: bad:
                   2873:        vm_map_unlock(srcmap);
                   2874: bad2:                  /* src already unlocked */
                   2875:        if (chain)
                   2876:                uvm_unmap_detach(chain,
                   2877:                    (flags & UVM_EXTRACT_QREF) ? AMAP_REFALL : 0);
1.144     yamt     2878:
1.204.2.1  yamt     2879:        uvm_map_check(srcmap, "map_extract src err leave");
                   2880:        uvm_map_check(dstmap, "map_extract dst err leave");
1.144     yamt     2881:
1.204.2.1  yamt     2882:        if ((flags & UVM_EXTRACT_RESERVED) == 0) {
                   2883:                uvm_unmap(dstmap, dstaddr, dstaddr+len);   /* ??? */
                   2884:        }
1.139     enami    2885:        return (error);
1.10      mrg      2886: }
                   2887:
                   2888: /* end of extraction functions */
1.1       mrg      2889:
                   2890: /*
                   2891:  * uvm_map_submap: punch down part of a map into a submap
                   2892:  *
                   2893:  * => only the kernel_map is allowed to be submapped
                   2894:  * => the purpose of submapping is to break up the locking granularity
                   2895:  *     of a larger map
                   2896:  * => the range specified must have been mapped previously with a uvm_map()
                   2897:  *     call [with uobj==NULL] to create a blank map entry in the main map.
                   2898:  *     [And it had better still be blank!]
                   2899:  * => maps which contain submaps should never be copied or forked.
1.98      chs      2900:  * => to remove a submap, use uvm_unmap() on the main map
1.1       mrg      2901:  *     and then uvm_map_deallocate() the submap.
                   2902:  * => main map must be unlocked.
                   2903:  * => submap must have been init'd and have a zero reference count.
                   2904:  *     [need not be locked as we don't actually reference it]
                   2905:  */
1.85      chs      2906:
1.10      mrg      2907: int
1.138     enami    2908: uvm_map_submap(struct vm_map *map, vaddr_t start, vaddr_t end,
                   2909:     struct vm_map *submap)
1.10      mrg      2910: {
1.99      chs      2911:        struct vm_map_entry *entry;
1.174     yamt     2912:        struct uvm_mapent_reservation umr;
1.94      chs      2913:        int error;
1.1       mrg      2914:
1.174     yamt     2915:        uvm_mapent_reserve(map, &umr, 2, 0);
                   2916:
1.10      mrg      2917:        vm_map_lock(map);
1.85      chs      2918:        VM_MAP_RANGE_CHECK(map, start, end);
1.1       mrg      2919:
1.10      mrg      2920:        if (uvm_map_lookup_entry(map, start, &entry)) {
1.174     yamt     2921:                UVM_MAP_CLIP_START(map, entry, start, &umr);
                   2922:                UVM_MAP_CLIP_END(map, entry, end, &umr);        /* to be safe */
1.94      chs      2923:        } else {
1.10      mrg      2924:                entry = NULL;
                   2925:        }
1.1       mrg      2926:
1.98      chs      2927:        if (entry != NULL &&
1.10      mrg      2928:            entry->start == start && entry->end == end &&
                   2929:            entry->object.uvm_obj == NULL && entry->aref.ar_amap == NULL &&
                   2930:            !UVM_ET_ISCOPYONWRITE(entry) && !UVM_ET_ISNEEDSCOPY(entry)) {
1.29      chuck    2931:                entry->etype |= UVM_ET_SUBMAP;
1.10      mrg      2932:                entry->object.sub_map = submap;
                   2933:                entry->offset = 0;
                   2934:                uvm_map_reference(submap);
1.94      chs      2935:                error = 0;
1.10      mrg      2936:        } else {
1.94      chs      2937:                error = EINVAL;
1.10      mrg      2938:        }
                   2939:        vm_map_unlock(map);
1.174     yamt     2940:
                   2941:        uvm_mapent_unreserve(map, &umr);
                   2942:
1.94      chs      2943:        return error;
1.1       mrg      2944: }
                   2945:
1.175     yamt     2946: /*
                   2947:  * uvm_map_setup_kernel: init in-kernel map
                   2948:  *
                   2949:  * => map must not be in service yet.
                   2950:  */
                   2951:
                   2952: void
                   2953: uvm_map_setup_kernel(struct vm_map_kernel *map,
1.199     christos 2954:     vaddr_t vmin, vaddr_t vmax, int flags)
1.175     yamt     2955: {
                   2956:
1.199     christos 2957:        uvm_map_setup(&map->vmk_map, vmin, vmax, flags);
1.204.2.4  yamt     2958:        callback_head_init(&map->vmk_reclaim_callback, IPL_VM);
1.175     yamt     2959:        LIST_INIT(&map->vmk_kentry_free);
                   2960:        map->vmk_merged_entries = NULL;
                   2961: }
                   2962:
1.1       mrg      2963:
                   2964: /*
                   2965:  * uvm_map_protect: change map protection
                   2966:  *
                   2967:  * => set_max means set max_protection.
                   2968:  * => map must be unlocked.
                   2969:  */
                   2970:
1.139     enami    2971: #define MASK(entry)    (UVM_ET_ISCOPYONWRITE(entry) ? \
1.36      mycroft  2972:                         ~VM_PROT_WRITE : VM_PROT_ALL)
1.1       mrg      2973:
1.10      mrg      2974: int
1.138     enami    2975: uvm_map_protect(struct vm_map *map, vaddr_t start, vaddr_t end,
1.204.2.3  yamt     2976:     vm_prot_t new_prot, bool set_max)
1.10      mrg      2977: {
1.99      chs      2978:        struct vm_map_entry *current, *entry;
1.94      chs      2979:        int error = 0;
1.10      mrg      2980:        UVMHIST_FUNC("uvm_map_protect"); UVMHIST_CALLED(maphist);
                   2981:        UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_prot=0x%x)",
1.85      chs      2982:                    map, start, end, new_prot);
                   2983:
1.10      mrg      2984:        vm_map_lock(map);
                   2985:        VM_MAP_RANGE_CHECK(map, start, end);
                   2986:        if (uvm_map_lookup_entry(map, start, &entry)) {
1.174     yamt     2987:                UVM_MAP_CLIP_START(map, entry, start, NULL);
1.10      mrg      2988:        } else {
                   2989:                entry = entry->next;
                   2990:        }
                   2991:
1.1       mrg      2992:        /*
1.10      mrg      2993:         * make a first pass to check for protection violations.
1.1       mrg      2994:         */
                   2995:
1.10      mrg      2996:        current = entry;
                   2997:        while ((current != &map->header) && (current->start < end)) {
1.65      thorpej  2998:                if (UVM_ET_ISSUBMAP(current)) {
1.94      chs      2999:                        error = EINVAL;
1.65      thorpej  3000:                        goto out;
                   3001:                }
1.10      mrg      3002:                if ((new_prot & current->max_protection) != new_prot) {
1.94      chs      3003:                        error = EACCES;
1.65      thorpej  3004:                        goto out;
1.112     thorpej  3005:                }
                   3006:                /*
                   3007:                 * Don't allow VM_PROT_EXECUTE to be set on entries that
                   3008:                 * point to vnodes that are associated with a NOEXEC file
                   3009:                 * system.
                   3010:                 */
                   3011:                if (UVM_ET_ISOBJ(current) &&
                   3012:                    UVM_OBJ_IS_VNODE(current->object.uvm_obj)) {
                   3013:                        struct vnode *vp =
                   3014:                            (struct vnode *) current->object.uvm_obj;
                   3015:
                   3016:                        if ((new_prot & VM_PROT_EXECUTE) != 0 &&
                   3017:                            (vp->v_mount->mnt_flag & MNT_NOEXEC) != 0) {
                   3018:                                error = EACCES;
                   3019:                                goto out;
                   3020:                        }
1.10      mrg      3021:                }
1.204.2.1  yamt     3022:
1.65      thorpej  3023:                current = current->next;
1.10      mrg      3024:        }
                   3025:
                   3026:        /* go back and fix up protections (no need to clip this time). */
                   3027:
                   3028:        current = entry;
                   3029:        while ((current != &map->header) && (current->start < end)) {
                   3030:                vm_prot_t old_prot;
1.85      chs      3031:
1.174     yamt     3032:                UVM_MAP_CLIP_END(map, current, end, NULL);
1.10      mrg      3033:                old_prot = current->protection;
                   3034:                if (set_max)
                   3035:                        current->protection =
                   3036:                            (current->max_protection = new_prot) & old_prot;
                   3037:                else
                   3038:                        current->protection = new_prot;
                   3039:
                   3040:                /*
1.98      chs      3041:                 * update physical map if necessary.  worry about copy-on-write
1.10      mrg      3042:                 * here -- CHECK THIS XXX
                   3043:                 */
                   3044:
                   3045:                if (current->protection != old_prot) {
1.29      chuck    3046:                        /* update pmap! */
                   3047:                        pmap_protect(map->pmap, current->start, current->end,
                   3048:                            current->protection & MASK(entry));
1.109     thorpej  3049:
                   3050:                        /*
                   3051:                         * If this entry points at a vnode, and the
                   3052:                         * protection includes VM_PROT_EXECUTE, mark
1.111     thorpej  3053:                         * the vnode as VEXECMAP.
1.109     thorpej  3054:                         */
                   3055:                        if (UVM_ET_ISOBJ(current)) {
                   3056:                                struct uvm_object *uobj =
                   3057:                                    current->object.uvm_obj;
                   3058:
                   3059:                                if (UVM_OBJ_IS_VNODE(uobj) &&
1.204.2.5  yamt     3060:                                    (current->protection & VM_PROT_EXECUTE)) {
1.204.2.7  yamt     3061:                                        mutex_enter(&uobj->vmobjlock);
1.110     thorpej  3062:                                        vn_markexec((struct vnode *) uobj);
1.204.2.7  yamt     3063:                                        mutex_exit(&uobj->vmobjlock);
1.204.2.5  yamt     3064:                                }
1.109     thorpej  3065:                        }
1.65      thorpej  3066:                }
1.10      mrg      3067:
1.65      thorpej  3068:                /*
                   3069:                 * If the map is configured to lock any future mappings,
                   3070:                 * wire this entry now if the old protection was VM_PROT_NONE
                   3071:                 * and the new protection is not VM_PROT_NONE.
                   3072:                 */
                   3073:
                   3074:                if ((map->flags & VM_MAP_WIREFUTURE) != 0 &&
                   3075:                    VM_MAPENT_ISWIRED(entry) == 0 &&
                   3076:                    old_prot == VM_PROT_NONE &&
                   3077:                    new_prot != VM_PROT_NONE) {
                   3078:                        if (uvm_map_pageable(map, entry->start,
1.204.2.3  yamt     3079:                            entry->end, false,
1.94      chs      3080:                            UVM_LK_ENTER|UVM_LK_EXIT) != 0) {
1.99      chs      3081:
1.65      thorpej  3082:                                /*
                   3083:                                 * If locking the entry fails, remember the
                   3084:                                 * error if it's the first one.  Note we
                   3085:                                 * still continue setting the protection in
1.94      chs      3086:                                 * the map, but will return the error
                   3087:                                 * condition regardless.
1.65      thorpej  3088:                                 *
                   3089:                                 * XXX Ignore what the actual error is,
                   3090:                                 * XXX just call it a resource shortage
                   3091:                                 * XXX so that it doesn't get confused
                   3092:                                 * XXX what uvm_map_protect() itself would
                   3093:                                 * XXX normally return.
                   3094:                                 */
1.99      chs      3095:
1.94      chs      3096:                                error = ENOMEM;
1.65      thorpej  3097:                        }
1.10      mrg      3098:                }
                   3099:                current = current->next;
                   3100:        }
1.105     chris    3101:        pmap_update(map->pmap);
1.85      chs      3102:
1.65      thorpej  3103:  out:
1.10      mrg      3104:        vm_map_unlock(map);
1.174     yamt     3105:
1.94      chs      3106:        UVMHIST_LOG(maphist, "<- done, error=%d",error,0,0,0);
                   3107:        return error;
1.1       mrg      3108: }
                   3109:
                   3110: #undef  MASK
                   3111:
1.98      chs      3112: /*
1.1       mrg      3113:  * uvm_map_inherit: set inheritance code for range of addrs in map.
                   3114:  *
                   3115:  * => map must be unlocked
                   3116:  * => note that the inherit code is used during a "fork".  see fork
                   3117:  *     code for details.
                   3118:  */
                   3119:
1.10      mrg      3120: int
1.138     enami    3121: uvm_map_inherit(struct vm_map *map, vaddr_t start, vaddr_t end,
                   3122:     vm_inherit_t new_inheritance)
1.10      mrg      3123: {
1.99      chs      3124:        struct vm_map_entry *entry, *temp_entry;
1.10      mrg      3125:        UVMHIST_FUNC("uvm_map_inherit"); UVMHIST_CALLED(maphist);
                   3126:        UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_inh=0x%x)",
                   3127:            map, start, end, new_inheritance);
                   3128:
                   3129:        switch (new_inheritance) {
1.80      wiz      3130:        case MAP_INHERIT_NONE:
                   3131:        case MAP_INHERIT_COPY:
                   3132:        case MAP_INHERIT_SHARE:
1.10      mrg      3133:                break;
                   3134:        default:
                   3135:                UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1.94      chs      3136:                return EINVAL;
1.10      mrg      3137:        }
1.1       mrg      3138:
1.10      mrg      3139:        vm_map_lock(map);
                   3140:        VM_MAP_RANGE_CHECK(map, start, end);
                   3141:        if (uvm_map_lookup_entry(map, start, &temp_entry)) {
                   3142:                entry = temp_entry;
1.174     yamt     3143:                UVM_MAP_CLIP_START(map, entry, start, NULL);
1.10      mrg      3144:        }  else {
                   3145:                entry = temp_entry->next;
                   3146:        }
                   3147:        while ((entry != &map->header) && (entry->start < end)) {
1.174     yamt     3148:                UVM_MAP_CLIP_END(map, entry, end, NULL);
1.10      mrg      3149:                entry->inheritance = new_inheritance;
                   3150:                entry = entry->next;
                   3151:        }
                   3152:        vm_map_unlock(map);
                   3153:        UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1.94      chs      3154:        return 0;
1.41      mrg      3155: }
                   3156:
1.98      chs      3157: /*
1.41      mrg      3158:  * uvm_map_advice: set advice code for range of addrs in map.
                   3159:  *
                   3160:  * => map must be unlocked
                   3161:  */
                   3162:
                   3163: int
1.138     enami    3164: uvm_map_advice(struct vm_map *map, vaddr_t start, vaddr_t end, int new_advice)
1.41      mrg      3165: {
1.99      chs      3166:        struct vm_map_entry *entry, *temp_entry;
1.41      mrg      3167:        UVMHIST_FUNC("uvm_map_advice"); UVMHIST_CALLED(maphist);
                   3168:        UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_adv=0x%x)",
                   3169:            map, start, end, new_advice);
                   3170:
                   3171:        vm_map_lock(map);
                   3172:        VM_MAP_RANGE_CHECK(map, start, end);
                   3173:        if (uvm_map_lookup_entry(map, start, &temp_entry)) {
                   3174:                entry = temp_entry;
1.174     yamt     3175:                UVM_MAP_CLIP_START(map, entry, start, NULL);
1.41      mrg      3176:        } else {
                   3177:                entry = temp_entry->next;
                   3178:        }
1.61      thorpej  3179:
                   3180:        /*
                   3181:         * XXXJRT: disallow holes?
                   3182:         */
                   3183:
1.41      mrg      3184:        while ((entry != &map->header) && (entry->start < end)) {
1.174     yamt     3185:                UVM_MAP_CLIP_END(map, entry, end, NULL);
1.41      mrg      3186:
                   3187:                switch (new_advice) {
                   3188:                case MADV_NORMAL:
                   3189:                case MADV_RANDOM:
                   3190:                case MADV_SEQUENTIAL:
                   3191:                        /* nothing special here */
                   3192:                        break;
                   3193:
                   3194:                default:
1.50      mrg      3195:                        vm_map_unlock(map);
1.41      mrg      3196:                        UVMHIST_LOG(maphist,"<- done (INVALID ARG)",0,0,0,0);
1.94      chs      3197:                        return EINVAL;
1.41      mrg      3198:                }
                   3199:                entry->advice = new_advice;
                   3200:                entry = entry->next;
                   3201:        }
                   3202:
                   3203:        vm_map_unlock(map);
                   3204:        UVMHIST_LOG(maphist,"<- done (OK)",0,0,0,0);
1.94      chs      3205:        return 0;
1.1       mrg      3206: }
                   3207:
                   3208: /*
                   3209:  * uvm_map_pageable: sets the pageability of a range in a map.
                   3210:  *
1.56      thorpej  3211:  * => wires map entries.  should not be used for transient page locking.
                   3212:  *     for that, use uvm_fault_wire()/uvm_fault_unwire() (see uvm_vslock()).
1.204.2.1  yamt     3213:  * => regions specified as not pageable require lock-down (wired) memory
1.1       mrg      3214:  *     and page tables.
1.59      thorpej  3215:  * => map must never be read-locked
1.204.2.3  yamt     3216:  * => if islocked is true, map is already write-locked
1.59      thorpej  3217:  * => we always unlock the map, since we must downgrade to a read-lock
                   3218:  *     to call uvm_fault_wire()
1.1       mrg      3219:  * => XXXCDC: check this and try and clean it up.
                   3220:  */
                   3221:
1.19      kleink   3222: int
1.138     enami    3223: uvm_map_pageable(struct vm_map *map, vaddr_t start, vaddr_t end,
1.204.2.3  yamt     3224:     bool new_pageable, int lockflags)
1.1       mrg      3225: {
1.99      chs      3226:        struct vm_map_entry *entry, *start_entry, *failed_entry;
1.10      mrg      3227:        int rv;
1.60      thorpej  3228: #ifdef DIAGNOSTIC
                   3229:        u_int timestamp_save;
                   3230: #endif
1.10      mrg      3231:        UVMHIST_FUNC("uvm_map_pageable"); UVMHIST_CALLED(maphist);
                   3232:        UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,new_pageable=0x%x)",
1.85      chs      3233:                    map, start, end, new_pageable);
                   3234:        KASSERT(map->flags & VM_MAP_PAGEABLE);
1.45      thorpej  3235:
1.64      thorpej  3236:        if ((lockflags & UVM_LK_ENTER) == 0)
1.59      thorpej  3237:                vm_map_lock(map);
1.10      mrg      3238:        VM_MAP_RANGE_CHECK(map, start, end);
                   3239:
1.98      chs      3240:        /*
1.10      mrg      3241:         * only one pageability change may take place at one time, since
                   3242:         * uvm_fault_wire assumes it will be called only once for each
                   3243:         * wiring/unwiring.  therefore, we have to make sure we're actually
                   3244:         * changing the pageability for the entire region.  we do so before
1.98      chs      3245:         * making any changes.
1.10      mrg      3246:         */
                   3247:
1.204.2.3  yamt     3248:        if (uvm_map_lookup_entry(map, start, &start_entry) == false) {
1.64      thorpej  3249:                if ((lockflags & UVM_LK_EXIT) == 0)
                   3250:                        vm_map_unlock(map);
1.85      chs      3251:
1.94      chs      3252:                UVMHIST_LOG(maphist,"<- done (fault)",0,0,0,0);
                   3253:                return EFAULT;
1.10      mrg      3254:        }
                   3255:        entry = start_entry;
                   3256:
1.98      chs      3257:        /*
1.100     wiz      3258:         * handle wiring and unwiring separately.
1.10      mrg      3259:         */
1.1       mrg      3260:
1.56      thorpej  3261:        if (new_pageable) {             /* unwire */
1.174     yamt     3262:                UVM_MAP_CLIP_START(map, entry, start, NULL);
1.85      chs      3263:
1.10      mrg      3264:                /*
                   3265:                 * unwiring.  first ensure that the range to be unwired is
1.98      chs      3266:                 * really wired down and that there are no holes.
1.10      mrg      3267:                 */
1.85      chs      3268:
1.10      mrg      3269:                while ((entry != &map->header) && (entry->start < end)) {
                   3270:                        if (entry->wired_count == 0 ||
                   3271:                            (entry->end < end &&
1.55      thorpej  3272:                             (entry->next == &map->header ||
                   3273:                              entry->next->start > entry->end))) {
1.64      thorpej  3274:                                if ((lockflags & UVM_LK_EXIT) == 0)
                   3275:                                        vm_map_unlock(map);
1.94      chs      3276:                                UVMHIST_LOG(maphist, "<- done (INVAL)",0,0,0,0);
                   3277:                                return EINVAL;
1.10      mrg      3278:                        }
                   3279:                        entry = entry->next;
                   3280:                }
                   3281:
1.98      chs      3282:                /*
1.56      thorpej  3283:                 * POSIX 1003.1b - a single munlock call unlocks a region,
                   3284:                 * regardless of the number of mlock calls made on that
                   3285:                 * region.
1.10      mrg      3286:                 */
1.85      chs      3287:
1.10      mrg      3288:                entry = start_entry;
                   3289:                while ((entry != &map->header) && (entry->start < end)) {
1.174     yamt     3290:                        UVM_MAP_CLIP_END(map, entry, end, NULL);
1.56      thorpej  3291:                        if (VM_MAPENT_ISWIRED(entry))
1.10      mrg      3292:                                uvm_map_entry_unwire(map, entry);
                   3293:                        entry = entry->next;
                   3294:                }
1.64      thorpej  3295:                if ((lockflags & UVM_LK_EXIT) == 0)
                   3296:                        vm_map_unlock(map);
1.10      mrg      3297:                UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
1.94      chs      3298:                return 0;
1.10      mrg      3299:        }
                   3300:
                   3301:        /*
                   3302:         * wire case: in two passes [XXXCDC: ugly block of code here]
                   3303:         *
                   3304:         * 1: holding the write lock, we create any anonymous maps that need
                   3305:         *    to be created.  then we clip each map entry to the region to
1.98      chs      3306:         *    be wired and increment its wiring count.
1.10      mrg      3307:         *
                   3308:         * 2: we downgrade to a read lock, and call uvm_fault_wire to fault
1.56      thorpej  3309:         *    in the pages for any newly wired area (wired_count == 1).
1.10      mrg      3310:         *
                   3311:         *    downgrading to a read lock for uvm_fault_wire avoids a possible
                   3312:         *    deadlock with another thread that may have faulted on one of
                   3313:         *    the pages to be wired (it would mark the page busy, blocking
                   3314:         *    us, then in turn block on the map lock that we hold).  because
                   3315:         *    of problems in the recursive lock package, we cannot upgrade
                   3316:         *    to a write lock in vm_map_lookup.  thus, any actions that
                   3317:         *    require the write lock must be done beforehand.  because we
                   3318:         *    keep the read lock on the map, the copy-on-write status of the
                   3319:         *    entries we modify here cannot change.
                   3320:         */
                   3321:
                   3322:        while ((entry != &map->header) && (entry->start < end)) {
1.55      thorpej  3323:                if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
1.85      chs      3324:
                   3325:                        /*
1.10      mrg      3326:                         * perform actions of vm_map_lookup that need the
                   3327:                         * write lock on the map: create an anonymous map
                   3328:                         * for a copy-on-write region, or an anonymous map
1.29      chuck    3329:                         * for a zero-fill region.  (XXXCDC: submap case
                   3330:                         * ok?)
1.10      mrg      3331:                         */
1.85      chs      3332:
1.29      chuck    3333:                        if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
1.98      chs      3334:                                if (UVM_ET_ISNEEDSCOPY(entry) &&
1.117     chs      3335:                                    ((entry->max_protection & VM_PROT_WRITE) ||
1.54      thorpej  3336:                                     (entry->object.uvm_obj == NULL))) {
1.204.2.1  yamt     3337:                                        amap_copy(map, entry, 0, start, end);
1.10      mrg      3338:                                        /* XXXCDC: wait OK? */
                   3339:                                }
                   3340:                        }
1.55      thorpej  3341:                }
1.174     yamt     3342:                UVM_MAP_CLIP_START(map, entry, start, NULL);
                   3343:                UVM_MAP_CLIP_END(map, entry, end, NULL);
1.10      mrg      3344:                entry->wired_count++;
                   3345:
                   3346:                /*
1.98      chs      3347:                 * Check for holes
1.10      mrg      3348:                 */
1.85      chs      3349:
1.54      thorpej  3350:                if (entry->protection == VM_PROT_NONE ||
                   3351:                    (entry->end < end &&
                   3352:                     (entry->next == &map->header ||
                   3353:                      entry->next->start > entry->end))) {
1.85      chs      3354:
1.10      mrg      3355:                        /*
                   3356:                         * found one.  amap creation actions do not need to
1.98      chs      3357:                         * be undone, but the wired counts need to be restored.
1.10      mrg      3358:                         */
1.85      chs      3359:
1.10      mrg      3360:                        while (entry != &map->header && entry->end > start) {
                   3361:                                entry->wired_count--;
                   3362:                                entry = entry->prev;
                   3363:                        }
1.64      thorpej  3364:                        if ((lockflags & UVM_LK_EXIT) == 0)
                   3365:                                vm_map_unlock(map);
1.10      mrg      3366:                        UVMHIST_LOG(maphist,"<- done (INVALID WIRE)",0,0,0,0);
1.94      chs      3367:                        return EINVAL;
1.10      mrg      3368:                }
                   3369:                entry = entry->next;
                   3370:        }
                   3371:
                   3372:        /*
                   3373:         * Pass 2.
                   3374:         */
1.51      thorpej  3375:
1.60      thorpej  3376: #ifdef DIAGNOSTIC
                   3377:        timestamp_save = map->timestamp;
                   3378: #endif
                   3379:        vm_map_busy(map);
1.204.2.7  yamt     3380:        vm_map_unlock(map);
1.10      mrg      3381:
                   3382:        rv = 0;
                   3383:        entry = start_entry;
                   3384:        while (entry != &map->header && entry->start < end) {
1.51      thorpej  3385:                if (entry->wired_count == 1) {
1.44      thorpej  3386:                        rv = uvm_fault_wire(map, entry->start, entry->end,
1.204.2.1  yamt     3387:                            entry->max_protection, 1);
1.10      mrg      3388:                        if (rv) {
1.94      chs      3389:
1.51      thorpej  3390:                                /*
                   3391:                                 * wiring failed.  break out of the loop.
                   3392:                                 * we'll clean up the map below, once we
                   3393:                                 * have a write lock again.
                   3394:                                 */
1.94      chs      3395:
1.51      thorpej  3396:                                break;
1.10      mrg      3397:                        }
                   3398:                }
                   3399:                entry = entry->next;
                   3400:        }
                   3401:
1.139     enami    3402:        if (rv) {       /* failed? */
1.85      chs      3403:
1.52      thorpej  3404:                /*
                   3405:                 * Get back to an exclusive (write) lock.
                   3406:                 */
1.85      chs      3407:
1.204.2.7  yamt     3408:                vm_map_lock(map);
1.60      thorpej  3409:                vm_map_unbusy(map);
                   3410:
                   3411: #ifdef DIAGNOSTIC
1.204.2.9! yamt     3412:                if (timestamp_save + 1 != map->timestamp)
1.60      thorpej  3413:                        panic("uvm_map_pageable: stale map");
                   3414: #endif
1.10      mrg      3415:
1.51      thorpej  3416:                /*
                   3417:                 * first drop the wiring count on all the entries
                   3418:                 * which haven't actually been wired yet.
                   3419:                 */
1.85      chs      3420:
1.54      thorpej  3421:                failed_entry = entry;
                   3422:                while (entry != &map->header && entry->start < end) {
1.51      thorpej  3423:                        entry->wired_count--;
1.54      thorpej  3424:                        entry = entry->next;
                   3425:                }
1.51      thorpej  3426:
                   3427:                /*
1.54      thorpej  3428:                 * now, unwire all the entries that were successfully
                   3429:                 * wired above.
1.51      thorpej  3430:                 */
1.85      chs      3431:
1.54      thorpej  3432:                entry = start_entry;
                   3433:                while (entry != failed_entry) {
                   3434:                        entry->wired_count--;
1.55      thorpej  3435:                        if (VM_MAPENT_ISWIRED(entry) == 0)
1.54      thorpej  3436:                                uvm_map_entry_unwire(map, entry);
                   3437:                        entry = entry->next;
                   3438:                }
1.64      thorpej  3439:                if ((lockflags & UVM_LK_EXIT) == 0)
                   3440:                        vm_map_unlock(map);
1.10      mrg      3441:                UVMHIST_LOG(maphist, "<- done (RV=%d)", rv,0,0,0);
1.139     enami    3442:                return (rv);
1.10      mrg      3443:        }
1.51      thorpej  3444:
1.64      thorpej  3445:        if ((lockflags & UVM_LK_EXIT) == 0) {
                   3446:                vm_map_unbusy(map);
                   3447:        } else {
1.85      chs      3448:
1.64      thorpej  3449:                /*
                   3450:                 * Get back to an exclusive (write) lock.
                   3451:                 */
1.85      chs      3452:
1.204.2.7  yamt     3453:                vm_map_lock(map);
1.64      thorpej  3454:                vm_map_unbusy(map);
                   3455:        }
                   3456:
1.10      mrg      3457:        UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
1.94      chs      3458:        return 0;
1.1       mrg      3459: }
                   3460:
                   3461: /*
1.54      thorpej  3462:  * uvm_map_pageable_all: special case of uvm_map_pageable - affects
                   3463:  * all mapped regions.
                   3464:  *
                   3465:  * => map must not be locked.
                   3466:  * => if no flags are specified, all regions are unwired.
                   3467:  * => XXXJRT: has some of the same problems as uvm_map_pageable() above.
                   3468:  */
                   3469:
                   3470: int
1.138     enami    3471: uvm_map_pageable_all(struct vm_map *map, int flags, vsize_t limit)
1.54      thorpej  3472: {
1.99      chs      3473:        struct vm_map_entry *entry, *failed_entry;
1.54      thorpej  3474:        vsize_t size;
                   3475:        int rv;
1.60      thorpej  3476: #ifdef DIAGNOSTIC
                   3477:        u_int timestamp_save;
                   3478: #endif
1.54      thorpej  3479:        UVMHIST_FUNC("uvm_map_pageable_all"); UVMHIST_CALLED(maphist);
                   3480:        UVMHIST_LOG(maphist,"(map=0x%x,flags=0x%x)", map, flags, 0, 0);
                   3481:
1.85      chs      3482:        KASSERT(map->flags & VM_MAP_PAGEABLE);
1.54      thorpej  3483:
                   3484:        vm_map_lock(map);
                   3485:
                   3486:        /*
                   3487:         * handle wiring and unwiring separately.
                   3488:         */
                   3489:
                   3490:        if (flags == 0) {                       /* unwire */
1.99      chs      3491:
1.54      thorpej  3492:                /*
1.56      thorpej  3493:                 * POSIX 1003.1b -- munlockall unlocks all regions,
                   3494:                 * regardless of how many times mlockall has been called.
1.54      thorpej  3495:                 */
1.99      chs      3496:
1.54      thorpej  3497:                for (entry = map->header.next; entry != &map->header;
                   3498:                     entry = entry->next) {
1.56      thorpej  3499:                        if (VM_MAPENT_ISWIRED(entry))
                   3500:                                uvm_map_entry_unwire(map, entry);
1.54      thorpej  3501:                }
1.204.2.4  yamt     3502:                map->flags &= ~VM_MAP_WIREFUTURE;
1.54      thorpej  3503:                vm_map_unlock(map);
                   3504:                UVMHIST_LOG(maphist,"<- done (OK UNWIRE)",0,0,0,0);
1.94      chs      3505:                return 0;
1.54      thorpej  3506:        }
                   3507:
                   3508:        if (flags & MCL_FUTURE) {
1.99      chs      3509:
1.54      thorpej  3510:                /*
                   3511:                 * must wire all future mappings; remember this.
                   3512:                 */
1.99      chs      3513:
1.204.2.4  yamt     3514:                map->flags |= VM_MAP_WIREFUTURE;
1.54      thorpej  3515:        }
                   3516:
                   3517:        if ((flags & MCL_CURRENT) == 0) {
1.99      chs      3518:
1.54      thorpej  3519:                /*
                   3520:                 * no more work to do!
                   3521:                 */
1.99      chs      3522:
1.54      thorpej  3523:                UVMHIST_LOG(maphist,"<- done (OK no wire)",0,0,0,0);
                   3524:                vm_map_unlock(map);
1.94      chs      3525:                return 0;
1.54      thorpej  3526:        }
                   3527:
                   3528:        /*
                   3529:         * wire case: in three passes [XXXCDC: ugly block of code here]
                   3530:         *
                   3531:         * 1: holding the write lock, count all pages mapped by non-wired
                   3532:         *    entries.  if this would cause us to go over our limit, we fail.
                   3533:         *
                   3534:         * 2: still holding the write lock, we create any anonymous maps that
                   3535:         *    need to be created.  then we increment its wiring count.
                   3536:         *
                   3537:         * 3: we downgrade to a read lock, and call uvm_fault_wire to fault
1.56      thorpej  3538:         *    in the pages for any newly wired area (wired_count == 1).
1.54      thorpej  3539:         *
                   3540:         *    downgrading to a read lock for uvm_fault_wire avoids a possible
                   3541:         *    deadlock with another thread that may have faulted on one of
                   3542:         *    the pages to be wired (it would mark the page busy, blocking
                   3543:         *    us, then in turn block on the map lock that we hold).  because
                   3544:         *    of problems in the recursive lock package, we cannot upgrade
                   3545:         *    to a write lock in vm_map_lookup.  thus, any actions that
                   3546:         *    require the write lock must be done beforehand.  because we
                   3547:         *    keep the read lock on the map, the copy-on-write status of the
                   3548:         *    entries we modify here cannot change.
                   3549:         */
                   3550:
                   3551:        for (size = 0, entry = map->header.next; entry != &map->header;
                   3552:             entry = entry->next) {
                   3553:                if (entry->protection != VM_PROT_NONE &&
1.55      thorpej  3554:                    VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
1.54      thorpej  3555:                        size += entry->end - entry->start;
                   3556:                }
                   3557:        }
                   3558:
                   3559:        if (atop(size) + uvmexp.wired > uvmexp.wiredmax) {
                   3560:                vm_map_unlock(map);
1.94      chs      3561:                return ENOMEM;
1.54      thorpej  3562:        }
                   3563:
                   3564:        if (limit != 0 &&
                   3565:            (size + ptoa(pmap_wired_count(vm_map_pmap(map))) > limit)) {
                   3566:                vm_map_unlock(map);
1.94      chs      3567:                return ENOMEM;
1.54      thorpej  3568:        }
                   3569:
                   3570:        /*
                   3571:         * Pass 2.
                   3572:         */
                   3573:
                   3574:        for (entry = map->header.next; entry != &map->header;
                   3575:             entry = entry->next) {
                   3576:                if (entry->protection == VM_PROT_NONE)
                   3577:                        continue;
1.55      thorpej  3578:                if (VM_MAPENT_ISWIRED(entry) == 0) { /* not already wired? */
1.99      chs      3579:
1.54      thorpej  3580:                        /*
                   3581:                         * perform actions of vm_map_lookup that need the
                   3582:                         * write lock on the map: create an anonymous map
                   3583:                         * for a copy-on-write region, or an anonymous map
                   3584:                         * for a zero-fill region.  (XXXCDC: submap case
                   3585:                         * ok?)
                   3586:                         */
1.99      chs      3587:
1.54      thorpej  3588:                        if (!UVM_ET_ISSUBMAP(entry)) {  /* not submap */
1.98      chs      3589:                                if (UVM_ET_ISNEEDSCOPY(entry) &&
1.117     chs      3590:                                    ((entry->max_protection & VM_PROT_WRITE) ||
1.54      thorpej  3591:                                     (entry->object.uvm_obj == NULL))) {
1.204.2.1  yamt     3592:                                        amap_copy(map, entry, 0, entry->start,
                   3593:                                            entry->end);
1.54      thorpej  3594:                                        /* XXXCDC: wait OK? */
                   3595:                                }
                   3596:                        }
1.55      thorpej  3597:                }
1.54      thorpej  3598:                entry->wired_count++;
                   3599:        }
                   3600:
                   3601:        /*
                   3602:         * Pass 3.
                   3603:         */
                   3604:
1.60      thorpej  3605: #ifdef DIAGNOSTIC
                   3606:        timestamp_save = map->timestamp;
                   3607: #endif
                   3608:        vm_map_busy(map);
1.204.2.7  yamt     3609:        vm_map_unlock(map);
1.54      thorpej  3610:
1.94      chs      3611:        rv = 0;
1.54      thorpej  3612:        for (entry = map->header.next; entry != &map->header;
                   3613:             entry = entry->next) {
                   3614:                if (entry->wired_count == 1) {
                   3615:                        rv = uvm_fault_wire(map, entry->start, entry->end,
1.204.2.1  yamt     3616:                            entry->max_protection, 1);
1.54      thorpej  3617:                        if (rv) {
1.99      chs      3618:
1.54      thorpej  3619:                                /*
                   3620:                                 * wiring failed.  break out of the loop.
                   3621:                                 * we'll clean up the map below, once we
                   3622:                                 * have a write lock again.
                   3623:                                 */
1.99      chs      3624:
1.54      thorpej  3625:                                break;
                   3626:                        }
                   3627:                }
                   3628:        }
                   3629:
1.99      chs      3630:        if (rv) {
                   3631:
1.54      thorpej  3632:                /*
                   3633:                 * Get back an exclusive (write) lock.
                   3634:                 */
1.99      chs      3635:
1.204.2.7  yamt     3636:                vm_map_lock(map);
1.60      thorpej  3637:                vm_map_unbusy(map);
                   3638:
                   3639: #ifdef DIAGNOSTIC
1.204.2.9! yamt     3640:                if (timestamp_save + 1 != map->timestamp)
1.60      thorpej  3641:                        panic("uvm_map_pageable_all: stale map");
                   3642: #endif
1.54      thorpej  3643:
                   3644:                /*
                   3645:                 * first drop the wiring count on all the entries
                   3646:                 * which haven't actually been wired yet.
1.67      thorpej  3647:                 *
                   3648:                 * Skip VM_PROT_NONE entries like we did above.
1.54      thorpej  3649:                 */
1.99      chs      3650:
1.54      thorpej  3651:                failed_entry = entry;
                   3652:                for (/* nothing */; entry != &map->header;
1.67      thorpej  3653:                     entry = entry->next) {
                   3654:                        if (entry->protection == VM_PROT_NONE)
                   3655:                                continue;
1.54      thorpej  3656:                        entry->wired_count--;
1.67      thorpej  3657:                }
1.54      thorpej  3658:
                   3659:                /*
                   3660:                 * now, unwire all the entries that were successfully
                   3661:                 * wired above.
1.67      thorpej  3662:                 *
                   3663:                 * Skip VM_PROT_NONE entries like we did above.
1.54      thorpej  3664:                 */
1.99      chs      3665:
1.54      thorpej  3666:                for (entry = map->header.next; entry != failed_entry;
                   3667:                     entry = entry->next) {
1.67      thorpej  3668:                        if (entry->protection == VM_PROT_NONE)
                   3669:                                continue;
1.54      thorpej  3670:                        entry->wired_count--;
1.67      thorpej  3671:                        if (VM_MAPENT_ISWIRED(entry))
1.54      thorpej  3672:                                uvm_map_entry_unwire(map, entry);
                   3673:                }
                   3674:                vm_map_unlock(map);
                   3675:                UVMHIST_LOG(maphist,"<- done (RV=%d)", rv,0,0,0);
                   3676:                return (rv);
                   3677:        }
                   3678:
1.60      thorpej  3679:        vm_map_unbusy(map);
1.54      thorpej  3680:
                   3681:        UVMHIST_LOG(maphist,"<- done (OK WIRE)",0,0,0,0);
1.94      chs      3682:        return 0;
1.54      thorpej  3683: }
                   3684:
                   3685: /*
1.61      thorpej  3686:  * uvm_map_clean: clean out a map range
1.1       mrg      3687:  *
                   3688:  * => valid flags:
1.61      thorpej  3689:  *   if (flags & PGO_CLEANIT): dirty pages are cleaned first
1.1       mrg      3690:  *   if (flags & PGO_SYNCIO): dirty pages are written synchronously
                   3691:  *   if (flags & PGO_DEACTIVATE): any cached pages are deactivated after clean
                   3692:  *   if (flags & PGO_FREE): any cached pages are freed after clean
                   3693:  * => returns an error if any part of the specified range isn't mapped
1.98      chs      3694:  * => never a need to flush amap layer since the anonymous memory has
1.61      thorpej  3695:  *     no permanent home, but may deactivate pages there
                   3696:  * => called from sys_msync() and sys_madvise()
1.1       mrg      3697:  * => caller must not write-lock map (read OK).
                   3698:  * => we may sleep while cleaning if SYNCIO [with map read-locked]
                   3699:  */
                   3700:
1.10      mrg      3701: int
1.138     enami    3702: uvm_map_clean(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
1.10      mrg      3703: {
1.99      chs      3704:        struct vm_map_entry *current, *entry;
1.61      thorpej  3705:        struct uvm_object *uobj;
                   3706:        struct vm_amap *amap;
                   3707:        struct vm_anon *anon;
                   3708:        struct vm_page *pg;
                   3709:        vaddr_t offset;
1.24      eeh      3710:        vsize_t size;
1.188     dbj      3711:        voff_t uoff;
1.106     chs      3712:        int error, refs;
1.10      mrg      3713:        UVMHIST_FUNC("uvm_map_clean"); UVMHIST_CALLED(maphist);
1.85      chs      3714:
1.10      mrg      3715:        UVMHIST_LOG(maphist,"(map=0x%x,start=0x%x,end=0x%x,flags=0x%x)",
1.85      chs      3716:                    map, start, end, flags);
                   3717:        KASSERT((flags & (PGO_FREE|PGO_DEACTIVATE)) !=
                   3718:                (PGO_FREE|PGO_DEACTIVATE));
1.61      thorpej  3719:
1.10      mrg      3720:        vm_map_lock_read(map);
                   3721:        VM_MAP_RANGE_CHECK(map, start, end);
1.204.2.3  yamt     3722:        if (uvm_map_lookup_entry(map, start, &entry) == false) {
1.10      mrg      3723:                vm_map_unlock_read(map);
1.94      chs      3724:                return EFAULT;
1.10      mrg      3725:        }
                   3726:
                   3727:        /*
1.186     chs      3728:         * Make a first pass to check for holes and wiring problems.
1.10      mrg      3729:         */
1.85      chs      3730:
1.10      mrg      3731:        for (current = entry; current->start < end; current = current->next) {
                   3732:                if (UVM_ET_ISSUBMAP(current)) {
                   3733:                        vm_map_unlock_read(map);
1.94      chs      3734:                        return EINVAL;
1.10      mrg      3735:                }
1.186     chs      3736:                if ((flags & PGO_FREE) != 0 && VM_MAPENT_ISWIRED(entry)) {
                   3737:                        vm_map_unlock_read(map);
                   3738:                        return EBUSY;
                   3739:                }
1.90      chs      3740:                if (end <= current->end) {
                   3741:                        break;
                   3742:                }
                   3743:                if (current->end != current->next->start) {
1.10      mrg      3744:                        vm_map_unlock_read(map);
1.94      chs      3745:                        return EFAULT;
1.10      mrg      3746:                }
                   3747:        }
                   3748:
1.94      chs      3749:        error = 0;
1.90      chs      3750:        for (current = entry; start < end; current = current->next) {
1.61      thorpej  3751:                amap = current->aref.ar_amap;   /* top layer */
                   3752:                uobj = current->object.uvm_obj; /* bottom layer */
1.85      chs      3753:                KASSERT(start >= current->start);
1.1       mrg      3754:
1.10      mrg      3755:                /*
1.61      thorpej  3756:                 * No amap cleaning necessary if:
                   3757:                 *
                   3758:                 *      (1) There's no amap.
                   3759:                 *
                   3760:                 *      (2) We're not deactivating or freeing pages.
1.10      mrg      3761:                 */
1.85      chs      3762:
1.90      chs      3763:                if (amap == NULL || (flags & (PGO_DEACTIVATE|PGO_FREE)) == 0)
1.61      thorpej  3764:                        goto flush_object;
                   3765:
                   3766:                amap_lock(amap);
                   3767:                offset = start - current->start;
1.90      chs      3768:                size = MIN(end, current->end) - start;
                   3769:                for ( ; size != 0; size -= PAGE_SIZE, offset += PAGE_SIZE) {
1.61      thorpej  3770:                        anon = amap_lookup(&current->aref, offset);
                   3771:                        if (anon == NULL)
                   3772:                                continue;
                   3773:
1.204.2.7  yamt     3774:                        mutex_enter(&anon->an_lock);
1.192     yamt     3775:                        pg = anon->an_page;
1.63      thorpej  3776:                        if (pg == NULL) {
1.204.2.7  yamt     3777:                                mutex_exit(&anon->an_lock);
1.63      thorpej  3778:                                continue;
                   3779:                        }
                   3780:
1.61      thorpej  3781:                        switch (flags & (PGO_CLEANIT|PGO_FREE|PGO_DEACTIVATE)) {
1.85      chs      3782:
1.61      thorpej  3783:                        /*
1.115     chs      3784:                         * In these first 3 cases, we just deactivate the page.
1.61      thorpej  3785:                         */
1.85      chs      3786:
1.61      thorpej  3787:                        case PGO_CLEANIT|PGO_FREE:
                   3788:                        case PGO_CLEANIT|PGO_DEACTIVATE:
                   3789:                        case PGO_DEACTIVATE:
1.68      thorpej  3790:  deactivate_it:
1.61      thorpej  3791:                                /*
1.115     chs      3792:                                 * skip the page if it's loaned or wired,
                   3793:                                 * since it shouldn't be on a paging queue
                   3794:                                 * at all in these cases.
1.61      thorpej  3795:                                 */
1.85      chs      3796:
1.204.2.7  yamt     3797:                                mutex_enter(&uvm_pageqlock);
1.115     chs      3798:                                if (pg->loan_count != 0 ||
                   3799:                                    pg->wire_count != 0) {
1.204.2.7  yamt     3800:                                        mutex_exit(&uvm_pageqlock);
                   3801:                                        mutex_exit(&anon->an_lock);
1.61      thorpej  3802:                                        continue;
                   3803:                                }
1.85      chs      3804:                                KASSERT(pg->uanon == anon);
1.61      thorpej  3805:                                uvm_pagedeactivate(pg);
1.204.2.7  yamt     3806:                                mutex_exit(&uvm_pageqlock);
                   3807:                                mutex_exit(&anon->an_lock);
1.61      thorpej  3808:                                continue;
                   3809:
                   3810:                        case PGO_FREE:
1.85      chs      3811:
1.68      thorpej  3812:                                /*
                   3813:                                 * If there are multiple references to
                   3814:                                 * the amap, just deactivate the page.
                   3815:                                 */
1.85      chs      3816:
1.68      thorpej  3817:                                if (amap_refs(amap) > 1)
                   3818:                                        goto deactivate_it;
                   3819:
1.115     chs      3820:                                /* skip the page if it's wired */
1.62      thorpej  3821:                                if (pg->wire_count != 0) {
1.204.2.7  yamt     3822:                                        mutex_exit(&anon->an_lock);
1.62      thorpej  3823:                                        continue;
                   3824:                                }
1.66      thorpej  3825:                                amap_unadd(&current->aref, offset);
1.61      thorpej  3826:                                refs = --anon->an_ref;
1.204.2.7  yamt     3827:                                mutex_exit(&anon->an_lock);
1.61      thorpej  3828:                                if (refs == 0)
                   3829:                                        uvm_anfree(anon);
                   3830:                                continue;
                   3831:                        }
                   3832:                }
                   3833:                amap_unlock(amap);
1.1       mrg      3834:
1.61      thorpej  3835:  flush_object:
1.10      mrg      3836:                /*
1.33      chuck    3837:                 * flush pages if we've got a valid backing object.
1.116     chs      3838:                 * note that we must always clean object pages before
                   3839:                 * freeing them since otherwise we could reveal stale
                   3840:                 * data from files.
1.10      mrg      3841:                 */
1.1       mrg      3842:
1.188     dbj      3843:                uoff = current->offset + (start - current->start);
1.90      chs      3844:                size = MIN(end, current->end) - start;
1.61      thorpej  3845:                if (uobj != NULL) {
1.204.2.7  yamt     3846:                        mutex_enter(&uobj->vmobjlock);
1.136     thorpej  3847:                        if (uobj->pgops->pgo_put != NULL)
1.188     dbj      3848:                                error = (uobj->pgops->pgo_put)(uobj, uoff,
                   3849:                                    uoff + size, flags | PGO_CLEANIT);
1.136     thorpej  3850:                        else
                   3851:                                error = 0;
1.10      mrg      3852:                }
                   3853:                start += size;
                   3854:        }
1.1       mrg      3855:        vm_map_unlock_read(map);
1.98      chs      3856:        return (error);
1.1       mrg      3857: }
                   3858:
                   3859:
                   3860: /*
                   3861:  * uvm_map_checkprot: check protection in map
                   3862:  *
                   3863:  * => must allow specified protection in a fully allocated region.
                   3864:  * => map must be read or write locked by caller.
                   3865:  */
                   3866:
1.204.2.3  yamt     3867: bool
1.138     enami    3868: uvm_map_checkprot(struct vm_map *map, vaddr_t start, vaddr_t end,
                   3869:     vm_prot_t protection)
1.10      mrg      3870: {
1.99      chs      3871:        struct vm_map_entry *entry;
                   3872:        struct vm_map_entry *tmp_entry;
1.10      mrg      3873:
1.94      chs      3874:        if (!uvm_map_lookup_entry(map, start, &tmp_entry)) {
1.204.2.3  yamt     3875:                return (false);
1.94      chs      3876:        }
                   3877:        entry = tmp_entry;
                   3878:        while (start < end) {
                   3879:                if (entry == &map->header) {
1.204.2.3  yamt     3880:                        return (false);
1.94      chs      3881:                }
1.85      chs      3882:
1.10      mrg      3883:                /*
                   3884:                 * no holes allowed
                   3885:                 */
                   3886:
1.94      chs      3887:                if (start < entry->start) {
1.204.2.3  yamt     3888:                        return (false);
1.94      chs      3889:                }
1.10      mrg      3890:
                   3891:                /*
                   3892:                 * check protection associated with entry
                   3893:                 */
1.1       mrg      3894:
1.94      chs      3895:                if ((entry->protection & protection) != protection) {
1.204.2.3  yamt     3896:                        return (false);
1.94      chs      3897:                }
                   3898:                start = entry->end;
                   3899:                entry = entry->next;
                   3900:        }
1.204.2.3  yamt     3901:        return (true);
1.1       mrg      3902: }
                   3903:
                   3904: /*
                   3905:  * uvmspace_alloc: allocate a vmspace structure.
                   3906:  *
                   3907:  * - structure includes vm_map and pmap
                   3908:  * - XXX: no locking on this structure
                   3909:  * - refcnt set to 1, rest must be init'd by caller
                   3910:  */
1.10      mrg      3911: struct vmspace *
1.199     christos 3912: uvmspace_alloc(vaddr_t vmin, vaddr_t vmax)
1.10      mrg      3913: {
                   3914:        struct vmspace *vm;
                   3915:        UVMHIST_FUNC("uvmspace_alloc"); UVMHIST_CALLED(maphist);
                   3916:
1.204.2.7  yamt     3917:        vm = pool_cache_get(&uvm_vmspace_cache, PR_WAITOK);
1.199     christos 3918:        uvmspace_init(vm, NULL, vmin, vmax);
1.15      thorpej  3919:        UVMHIST_LOG(maphist,"<- done (vm=0x%x)", vm,0,0,0);
                   3920:        return (vm);
                   3921: }
                   3922:
                   3923: /*
                   3924:  * uvmspace_init: initialize a vmspace structure.
                   3925:  *
                   3926:  * - XXX: no locking on this structure
1.132     matt     3927:  * - refcnt set to 1, rest must be init'd by caller
1.15      thorpej  3928:  */
                   3929: void
1.199     christos 3930: uvmspace_init(struct vmspace *vm, struct pmap *pmap, vaddr_t vmin, vaddr_t vmax)
1.15      thorpej  3931: {
                   3932:        UVMHIST_FUNC("uvmspace_init"); UVMHIST_CALLED(maphist);
                   3933:
1.23      perry    3934:        memset(vm, 0, sizeof(*vm));
1.199     christos 3935:        uvm_map_setup(&vm->vm_map, vmin, vmax, VM_MAP_PAGEABLE
1.131     atatat   3936: #ifdef __USING_TOPDOWN_VM
                   3937:            | VM_MAP_TOPDOWN
                   3938: #endif
                   3939:            );
1.15      thorpej  3940:        if (pmap)
                   3941:                pmap_reference(pmap);
                   3942:        else
                   3943:                pmap = pmap_create();
                   3944:        vm->vm_map.pmap = pmap;
1.10      mrg      3945:        vm->vm_refcnt = 1;
1.15      thorpej  3946:        UVMHIST_LOG(maphist,"<- done",0,0,0,0);
1.1       mrg      3947: }
                   3948:
                   3949: /*
1.168     junyoung 3950:  * uvmspace_share: share a vmspace between two processes
1.1       mrg      3951:  *
                   3952:  * - used for vfork, threads(?)
                   3953:  */
                   3954:
1.10      mrg      3955: void
1.138     enami    3956: uvmspace_share(struct proc *p1, struct proc *p2)
1.1       mrg      3957: {
1.139     enami    3958:
1.204.2.1  yamt     3959:        uvmspace_addref(p1->p_vmspace);
1.10      mrg      3960:        p2->p_vmspace = p1->p_vmspace;
1.1       mrg      3961: }
                   3962:
                   3963: /*
                   3964:  * uvmspace_unshare: ensure that process "p" has its own, unshared, vmspace
                   3965:  *
                   3966:  * - XXX: no locking on vmspace
                   3967:  */
                   3968:
1.10      mrg      3969: void
1.138     enami    3970: uvmspace_unshare(struct lwp *l)
1.10      mrg      3971: {
1.128     thorpej  3972:        struct proc *p = l->l_proc;
1.10      mrg      3973:        struct vmspace *nvm, *ovm = p->p_vmspace;
1.85      chs      3974:
1.10      mrg      3975:        if (ovm->vm_refcnt == 1)
                   3976:                /* nothing to do: vmspace isn't shared in the first place */
                   3977:                return;
1.85      chs      3978:
1.10      mrg      3979:        /* make a new vmspace, still holding old one */
                   3980:        nvm = uvmspace_fork(ovm);
                   3981:
1.128     thorpej  3982:        pmap_deactivate(l);             /* unbind old vmspace */
1.98      chs      3983:        p->p_vmspace = nvm;
1.128     thorpej  3984:        pmap_activate(l);               /* switch to new vmspace */
1.13      thorpej  3985:
1.10      mrg      3986:        uvmspace_free(ovm);             /* drop reference to old vmspace */
1.1       mrg      3987: }
                   3988:
                   3989: /*
                   3990:  * uvmspace_exec: the process wants to exec a new program
                   3991:  */
                   3992:
1.10      mrg      3993: void
1.138     enami    3994: uvmspace_exec(struct lwp *l, vaddr_t start, vaddr_t end)
1.1       mrg      3995: {
1.128     thorpej  3996:        struct proc *p = l->l_proc;
1.10      mrg      3997:        struct vmspace *nvm, *ovm = p->p_vmspace;
1.99      chs      3998:        struct vm_map *map = &ovm->vm_map;
1.1       mrg      3999:
1.71      chs      4000: #ifdef __sparc__
1.10      mrg      4001:        /* XXX cgd 960926: the sparc #ifdef should be a MD hook */
1.128     thorpej  4002:        kill_user_windows(l);   /* before stack addresses go away */
1.1       mrg      4003: #endif
                   4004:
1.10      mrg      4005:        /*
                   4006:         * see if more than one process is using this vmspace...
                   4007:         */
1.1       mrg      4008:
1.10      mrg      4009:        if (ovm->vm_refcnt == 1) {
1.1       mrg      4010:
1.10      mrg      4011:                /*
                   4012:                 * if p is the only process using its vmspace then we can safely
                   4013:                 * recycle that vmspace for the program that is being exec'd.
                   4014:                 */
1.1       mrg      4015:
                   4016: #ifdef SYSVSHM
1.10      mrg      4017:                /*
                   4018:                 * SYSV SHM semantics require us to kill all segments on an exec
                   4019:                 */
1.99      chs      4020:
1.10      mrg      4021:                if (ovm->vm_shm)
                   4022:                        shmexit(ovm);
                   4023: #endif
1.54      thorpej  4024:
                   4025:                /*
                   4026:                 * POSIX 1003.1b -- "lock future mappings" is revoked
                   4027:                 * when a process execs another program image.
                   4028:                 */
1.99      chs      4029:
1.204.2.4  yamt     4030:                map->flags &= ~VM_MAP_WIREFUTURE;
1.10      mrg      4031:
                   4032:                /*
                   4033:                 * now unmap the old program
                   4034:                 */
1.99      chs      4035:
1.120     chs      4036:                pmap_remove_all(map->pmap);
1.184     chs      4037:                uvm_unmap(map, vm_map_min(map), vm_map_max(map));
1.144     yamt     4038:                KASSERT(map->header.prev == &map->header);
                   4039:                KASSERT(map->nentries == 0);
1.93      eeh      4040:
                   4041:                /*
                   4042:                 * resize the map
                   4043:                 */
1.99      chs      4044:
1.184     chs      4045:                vm_map_setmin(map, start);
                   4046:                vm_map_setmax(map, end);
1.10      mrg      4047:        } else {
                   4048:
                   4049:                /*
                   4050:                 * p's vmspace is being shared, so we can't reuse it for p since
                   4051:                 * it is still being used for others.   allocate a new vmspace
                   4052:                 * for p
                   4053:                 */
1.99      chs      4054:
1.101     chs      4055:                nvm = uvmspace_alloc(start, end);
1.1       mrg      4056:
1.10      mrg      4057:                /*
                   4058:                 * install new vmspace and drop our ref to the old one.
                   4059:                 */
                   4060:
1.128     thorpej  4061:                pmap_deactivate(l);
1.10      mrg      4062:                p->p_vmspace = nvm;
1.128     thorpej  4063:                pmap_activate(l);
1.13      thorpej  4064:
1.10      mrg      4065:                uvmspace_free(ovm);
                   4066:        }
1.1       mrg      4067: }
                   4068:
                   4069: /*
1.204.2.1  yamt     4070:  * uvmspace_addref: add a referece to a vmspace.
                   4071:  */
                   4072:
                   4073: void
                   4074: uvmspace_addref(struct vmspace *vm)
                   4075: {
                   4076:        struct vm_map *map = &vm->vm_map;
                   4077:
                   4078:        KASSERT((map->flags & VM_MAP_DYING) == 0);
                   4079:
1.204.2.4  yamt     4080:        mutex_enter(&map->misc_lock);
1.204.2.1  yamt     4081:        KASSERT(vm->vm_refcnt > 0);
                   4082:        vm->vm_refcnt++;
1.204.2.4  yamt     4083:        mutex_exit(&map->misc_lock);
1.204.2.1  yamt     4084: }
                   4085:
                   4086: /*
1.1       mrg      4087:  * uvmspace_free: free a vmspace data structure
                   4088:  */
                   4089:
1.10      mrg      4090: void
1.138     enami    4091: uvmspace_free(struct vmspace *vm)
1.1       mrg      4092: {
1.99      chs      4093:        struct vm_map_entry *dead_entries;
1.171     pk       4094:        struct vm_map *map = &vm->vm_map;
1.172     he       4095:        int n;
                   4096:
1.10      mrg      4097:        UVMHIST_FUNC("uvmspace_free"); UVMHIST_CALLED(maphist);
1.1       mrg      4098:
1.10      mrg      4099:        UVMHIST_LOG(maphist,"(vm=0x%x) ref=%d", vm, vm->vm_refcnt,0,0);
1.204.2.4  yamt     4100:        mutex_enter(&map->misc_lock);
1.172     he       4101:        n = --vm->vm_refcnt;
1.204.2.4  yamt     4102:        mutex_exit(&map->misc_lock);
1.171     pk       4103:        if (n > 0)
1.120     chs      4104:                return;
1.99      chs      4105:
1.120     chs      4106:        /*
                   4107:         * at this point, there should be no other references to the map.
                   4108:         * delete all of the mappings, then destroy the pmap.
                   4109:         */
1.99      chs      4110:
1.120     chs      4111:        map->flags |= VM_MAP_DYING;
                   4112:        pmap_remove_all(map->pmap);
1.92      thorpej  4113: #ifdef SYSVSHM
1.120     chs      4114:        /* Get rid of any SYSV shared memory segments. */
                   4115:        if (vm->vm_shm != NULL)
                   4116:                shmexit(vm);
1.92      thorpej  4117: #endif
1.120     chs      4118:        if (map->nentries) {
1.184     chs      4119:                uvm_unmap_remove(map, vm_map_min(map), vm_map_max(map),
1.187     yamt     4120:                    &dead_entries, NULL, 0);
1.120     chs      4121:                if (dead_entries != NULL)
                   4122:                        uvm_unmap_detach(dead_entries, 0);
1.10      mrg      4123:        }
1.146     yamt     4124:        KASSERT(map->nentries == 0);
                   4125:        KASSERT(map->size == 0);
1.204.2.4  yamt     4126:        mutex_destroy(&map->misc_lock);
                   4127:        mutex_destroy(&map->mutex);
                   4128:        rw_destroy(&map->lock);
1.120     chs      4129:        pmap_destroy(map->pmap);
1.204.2.7  yamt     4130:        pool_cache_put(&uvm_vmspace_cache, vm);
1.1       mrg      4131: }
                   4132:
                   4133: /*
                   4134:  *   F O R K   -   m a i n   e n t r y   p o i n t
                   4135:  */
                   4136: /*
                   4137:  * uvmspace_fork: fork a process' main map
                   4138:  *
                   4139:  * => create a new vmspace for child process from parent.
                   4140:  * => parent's map must not be locked.
                   4141:  */
                   4142:
1.10      mrg      4143: struct vmspace *
1.138     enami    4144: uvmspace_fork(struct vmspace *vm1)
1.10      mrg      4145: {
                   4146:        struct vmspace *vm2;
1.99      chs      4147:        struct vm_map *old_map = &vm1->vm_map;
                   4148:        struct vm_map *new_map;
                   4149:        struct vm_map_entry *old_entry;
                   4150:        struct vm_map_entry *new_entry;
1.10      mrg      4151:        UVMHIST_FUNC("uvmspace_fork"); UVMHIST_CALLED(maphist);
1.1       mrg      4152:
1.10      mrg      4153:        vm_map_lock(old_map);
1.1       mrg      4154:
1.184     chs      4155:        vm2 = uvmspace_alloc(vm_map_min(old_map), vm_map_max(old_map));
1.23      perry    4156:        memcpy(&vm2->vm_startcopy, &vm1->vm_startcopy,
1.204.2.4  yamt     4157:            (char *) (vm1 + 1) - (char *) &vm1->vm_startcopy);
1.10      mrg      4158:        new_map = &vm2->vm_map;           /* XXX */
                   4159:
                   4160:        old_entry = old_map->header.next;
1.162     pooka    4161:        new_map->size = old_map->size;
1.10      mrg      4162:
                   4163:        /*
                   4164:         * go entry-by-entry
                   4165:         */
1.1       mrg      4166:
1.10      mrg      4167:        while (old_entry != &old_map->header) {
1.1       mrg      4168:
1.10      mrg      4169:                /*
                   4170:                 * first, some sanity checks on the old entry
                   4171:                 */
1.99      chs      4172:
1.94      chs      4173:                KASSERT(!UVM_ET_ISSUBMAP(old_entry));
                   4174:                KASSERT(UVM_ET_ISCOPYONWRITE(old_entry) ||
                   4175:                        !UVM_ET_ISNEEDSCOPY(old_entry));
1.1       mrg      4176:
1.10      mrg      4177:                switch (old_entry->inheritance) {
1.80      wiz      4178:                case MAP_INHERIT_NONE:
1.99      chs      4179:
1.10      mrg      4180:                        /*
1.162     pooka    4181:                         * drop the mapping, modify size
1.10      mrg      4182:                         */
1.162     pooka    4183:                        new_map->size -= old_entry->end - old_entry->start;
1.10      mrg      4184:                        break;
                   4185:
1.80      wiz      4186:                case MAP_INHERIT_SHARE:
1.99      chs      4187:
1.10      mrg      4188:                        /*
                   4189:                         * share the mapping: this means we want the old and
                   4190:                         * new entries to share amaps and backing objects.
                   4191:                         */
                   4192:                        /*
                   4193:                         * if the old_entry needs a new amap (due to prev fork)
                   4194:                         * then we need to allocate it now so that we have
                   4195:                         * something we own to share with the new_entry.   [in
                   4196:                         * other words, we need to clear needs_copy]
                   4197:                         */
                   4198:
                   4199:                        if (UVM_ET_ISNEEDSCOPY(old_entry)) {
                   4200:                                /* get our own amap, clears needs_copy */
1.204.2.1  yamt     4201:                                amap_copy(old_map, old_entry, AMAP_COPY_NOCHUNK,
1.98      chs      4202:                                    0, 0);
1.10      mrg      4203:                                /* XXXCDC: WAITOK??? */
                   4204:                        }
                   4205:
1.126     bouyer   4206:                        new_entry = uvm_mapent_alloc(new_map, 0);
1.10      mrg      4207:                        /* old_entry -> new_entry */
                   4208:                        uvm_mapent_copy(old_entry, new_entry);
                   4209:
                   4210:                        /* new pmap has nothing wired in it */
                   4211:                        new_entry->wired_count = 0;
                   4212:
                   4213:                        /*
1.29      chuck    4214:                         * gain reference to object backing the map (can't
                   4215:                         * be a submap, already checked this case).
1.10      mrg      4216:                         */
1.99      chs      4217:
1.10      mrg      4218:                        if (new_entry->aref.ar_amap)
1.85      chs      4219:                                uvm_map_reference_amap(new_entry, AMAP_SHARED);
1.10      mrg      4220:
                   4221:                        if (new_entry->object.uvm_obj &&
                   4222:                            new_entry->object.uvm_obj->pgops->pgo_reference)
                   4223:                                new_entry->object.uvm_obj->
                   4224:                                    pgops->pgo_reference(
                   4225:                                        new_entry->object.uvm_obj);
                   4226:
                   4227:                        /* insert entry at end of new_map's entry list */
                   4228:                        uvm_map_entry_link(new_map, new_map->header.prev,
                   4229:                            new_entry);
                   4230:
                   4231:                        break;
                   4232:
1.80      wiz      4233:                case MAP_INHERIT_COPY:
1.10      mrg      4234:
                   4235:                        /*
                   4236:                         * copy-on-write the mapping (using mmap's
                   4237:                         * MAP_PRIVATE semantics)
1.29      chuck    4238:                         *
1.98      chs      4239:                         * allocate new_entry, adjust reference counts.
1.29      chuck    4240:                         * (note that new references are read-only).
1.10      mrg      4241:                         */
                   4242:
1.126     bouyer   4243:                        new_entry = uvm_mapent_alloc(new_map, 0);
1.10      mrg      4244:                        /* old_entry -> new_entry */
                   4245:                        uvm_mapent_copy(old_entry, new_entry);
                   4246:
                   4247:                        if (new_entry->aref.ar_amap)
1.85      chs      4248:                                uvm_map_reference_amap(new_entry, 0);
1.10      mrg      4249:
                   4250:                        if (new_entry->object.uvm_obj &&
                   4251:                            new_entry->object.uvm_obj->pgops->pgo_reference)
                   4252:                                new_entry->object.uvm_obj->pgops->pgo_reference
                   4253:                                    (new_entry->object.uvm_obj);
                   4254:
                   4255:                        /* new pmap has nothing wired in it */
                   4256:                        new_entry->wired_count = 0;
                   4257:
                   4258:                        new_entry->etype |=
                   4259:                            (UVM_ET_COPYONWRITE|UVM_ET_NEEDSCOPY);
                   4260:                        uvm_map_entry_link(new_map, new_map->header.prev,
                   4261:                            new_entry);
1.85      chs      4262:
1.14      chuck    4263:                        /*
1.10      mrg      4264:                         * the new entry will need an amap.  it will either
                   4265:                         * need to be copied from the old entry or created
1.14      chuck    4266:                         * from scratch (if the old entry does not have an
                   4267:                         * amap).  can we defer this process until later
                   4268:                         * (by setting "needs_copy") or do we need to copy
                   4269:                         * the amap now?
1.10      mrg      4270:                         *
1.14      chuck    4271:                         * we must copy the amap now if any of the following
1.10      mrg      4272:                         * conditions hold:
1.14      chuck    4273:                         * 1. the old entry has an amap and that amap is
                   4274:                         *    being shared.  this means that the old (parent)
1.98      chs      4275:                         *    process is sharing the amap with another
1.14      chuck    4276:                         *    process.  if we do not clear needs_copy here
                   4277:                         *    we will end up in a situation where both the
                   4278:                         *    parent and child process are refering to the
1.98      chs      4279:                         *    same amap with "needs_copy" set.  if the
1.14      chuck    4280:                         *    parent write-faults, the fault routine will
                   4281:                         *    clear "needs_copy" in the parent by allocating
1.98      chs      4282:                         *    a new amap.   this is wrong because the
1.14      chuck    4283:                         *    parent is supposed to be sharing the old amap
                   4284:                         *    and the new amap will break that.
1.10      mrg      4285:                         *
1.14      chuck    4286:                         * 2. if the old entry has an amap and a non-zero
                   4287:                         *    wire count then we are going to have to call
1.98      chs      4288:                         *    amap_cow_now to avoid page faults in the
1.14      chuck    4289:                         *    parent process.   since amap_cow_now requires
                   4290:                         *    "needs_copy" to be clear we might as well
                   4291:                         *    clear it here as well.
1.10      mrg      4292:                         *
                   4293:                         */
                   4294:
1.14      chuck    4295:                        if (old_entry->aref.ar_amap != NULL) {
1.99      chs      4296:                                if ((amap_flags(old_entry->aref.ar_amap) &
                   4297:                                     AMAP_SHARED) != 0 ||
                   4298:                                    VM_MAPENT_ISWIRED(old_entry)) {
                   4299:
1.204.2.1  yamt     4300:                                        amap_copy(new_map, new_entry,
                   4301:                                            AMAP_COPY_NOCHUNK, 0, 0);
1.99      chs      4302:                                        /* XXXCDC: M_WAITOK ... ok? */
                   4303:                                }
1.10      mrg      4304:                        }
1.85      chs      4305:
1.10      mrg      4306:                        /*
1.14      chuck    4307:                         * if the parent's entry is wired down, then the
                   4308:                         * parent process does not want page faults on
                   4309:                         * access to that memory.  this means that we
                   4310:                         * cannot do copy-on-write because we can't write
                   4311:                         * protect the old entry.   in this case we
                   4312:                         * resolve all copy-on-write faults now, using
                   4313:                         * amap_cow_now.   note that we have already
                   4314:                         * allocated any needed amap (above).
1.10      mrg      4315:                         */
                   4316:
1.55      thorpej  4317:                        if (VM_MAPENT_ISWIRED(old_entry)) {
1.1       mrg      4318:
1.98      chs      4319:                          /*
1.14      chuck    4320:                           * resolve all copy-on-write faults now
1.98      chs      4321:                           * (note that there is nothing to do if
1.14      chuck    4322:                           * the old mapping does not have an amap).
                   4323:                           */
                   4324:                          if (old_entry->aref.ar_amap)
                   4325:                            amap_cow_now(new_map, new_entry);
                   4326:
1.98      chs      4327:                        } else {
1.14      chuck    4328:
                   4329:                          /*
                   4330:                           * setup mappings to trigger copy-on-write faults
                   4331:                           * we must write-protect the parent if it has
                   4332:                           * an amap and it is not already "needs_copy"...
                   4333:                           * if it is already "needs_copy" then the parent
                   4334:                           * has already been write-protected by a previous
                   4335:                           * fork operation.
                   4336:                           */
                   4337:
1.113     chs      4338:                          if (old_entry->aref.ar_amap &&
                   4339:                              !UVM_ET_ISNEEDSCOPY(old_entry)) {
1.14      chuck    4340:                              if (old_entry->max_protection & VM_PROT_WRITE) {
                   4341:                                pmap_protect(old_map->pmap,
                   4342:                                             old_entry->start,
                   4343:                                             old_entry->end,
                   4344:                                             old_entry->protection &
                   4345:                                             ~VM_PROT_WRITE);
1.105     chris    4346:                                pmap_update(old_map->pmap);
1.14      chuck    4347:                              }
                   4348:                              old_entry->etype |= UVM_ET_NEEDSCOPY;
                   4349:                          }
1.10      mrg      4350:                        }
                   4351:                        break;
1.14      chuck    4352:                }  /* end of switch statement */
1.10      mrg      4353:                old_entry = old_entry->next;
1.1       mrg      4354:        }
                   4355:
1.98      chs      4356:        vm_map_unlock(old_map);
1.1       mrg      4357:
                   4358: #ifdef SYSVSHM
1.10      mrg      4359:        if (vm1->vm_shm)
                   4360:                shmfork(vm1, vm2);
1.39      thorpej  4361: #endif
                   4362:
                   4363: #ifdef PMAP_FORK
                   4364:        pmap_fork(vm1->vm_map.pmap, vm2->vm_map.pmap);
1.1       mrg      4365: #endif
                   4366:
1.10      mrg      4367:        UVMHIST_LOG(maphist,"<- done",0,0,0,0);
1.139     enami    4368:        return (vm2);
1.1       mrg      4369: }
                   4370:
                   4371:
1.174     yamt     4372: /*
                   4373:  * in-kernel map entry allocation.
                   4374:  */
                   4375:
                   4376: struct uvm_kmapent_hdr {
                   4377:        LIST_ENTRY(uvm_kmapent_hdr) ukh_listq;
                   4378:        int ukh_nused;
                   4379:        struct vm_map_entry *ukh_freelist;
                   4380:        struct vm_map *ukh_map;
1.178     yamt     4381:        struct vm_map_entry ukh_entries[0];
1.174     yamt     4382: };
                   4383:
                   4384: #define        UVM_KMAPENT_CHUNK                               \
                   4385:        ((PAGE_SIZE - sizeof(struct uvm_kmapent_hdr))   \
                   4386:        / sizeof(struct vm_map_entry))
                   4387:
                   4388: #define        UVM_KHDR_FIND(entry)    \
                   4389:        ((struct uvm_kmapent_hdr *)(((vaddr_t)entry) & ~PAGE_MASK))
                   4390:
                   4391:
1.201     dsl      4392: #ifdef DIAGNOSTIC
1.203     thorpej  4393: static struct vm_map *
1.174     yamt     4394: uvm_kmapent_map(struct vm_map_entry *entry)
                   4395: {
                   4396:        const struct uvm_kmapent_hdr *ukh;
                   4397:
                   4398:        ukh = UVM_KHDR_FIND(entry);
                   4399:        return ukh->ukh_map;
                   4400: }
1.201     dsl      4401: #endif
1.174     yamt     4402:
1.204.2.1  yamt     4403: static inline struct vm_map_entry *
1.174     yamt     4404: uvm_kmapent_get(struct uvm_kmapent_hdr *ukh)
                   4405: {
                   4406:        struct vm_map_entry *entry;
                   4407:
                   4408:        KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
                   4409:        KASSERT(ukh->ukh_nused >= 0);
                   4410:
                   4411:        entry = ukh->ukh_freelist;
                   4412:        if (entry) {
                   4413:                KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
                   4414:                    == UVM_MAP_KERNEL);
                   4415:                ukh->ukh_freelist = entry->next;
                   4416:                ukh->ukh_nused++;
                   4417:                KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
                   4418:        } else {
                   4419:                KASSERT(ukh->ukh_nused == UVM_KMAPENT_CHUNK);
                   4420:        }
                   4421:
                   4422:        return entry;
                   4423: }
                   4424:
1.204.2.1  yamt     4425: static inline void
1.174     yamt     4426: uvm_kmapent_put(struct uvm_kmapent_hdr *ukh, struct vm_map_entry *entry)
                   4427: {
                   4428:
                   4429:        KASSERT((entry->flags & (UVM_MAP_KERNEL | UVM_MAP_KMAPENT))
                   4430:            == UVM_MAP_KERNEL);
                   4431:        KASSERT(ukh->ukh_nused <= UVM_KMAPENT_CHUNK);
                   4432:        KASSERT(ukh->ukh_nused > 0);
                   4433:        KASSERT(ukh->ukh_freelist != NULL ||
                   4434:            ukh->ukh_nused == UVM_KMAPENT_CHUNK);
                   4435:        KASSERT(ukh->ukh_freelist == NULL ||
                   4436:            ukh->ukh_nused < UVM_KMAPENT_CHUNK);
                   4437:
                   4438:        ukh->ukh_nused--;
                   4439:        entry->next = ukh->ukh_freelist;
                   4440:        ukh->ukh_freelist = entry;
                   4441: }
                   4442:
                   4443: /*
                   4444:  * uvm_kmapent_alloc: allocate a map entry for in-kernel map
                   4445:  */
                   4446:
                   4447: static struct vm_map_entry *
                   4448: uvm_kmapent_alloc(struct vm_map *map, int flags)
                   4449: {
                   4450:        struct vm_page *pg;
                   4451:        struct uvm_map_args args;
                   4452:        struct uvm_kmapent_hdr *ukh;
                   4453:        struct vm_map_entry *entry;
                   4454:        uvm_flag_t mapflags = UVM_MAPFLAG(UVM_PROT_ALL, UVM_PROT_ALL,
                   4455:            UVM_INH_NONE, UVM_ADV_RANDOM, flags | UVM_FLAG_NOMERGE);
                   4456:        vaddr_t va;
                   4457:        int error;
                   4458:        int i;
                   4459:
                   4460:        KDASSERT(UVM_KMAPENT_CHUNK > 2);
                   4461:        KDASSERT(kernel_map != NULL);
                   4462:        KASSERT(vm_map_pmap(map) == pmap_kernel());
                   4463:
1.204.2.1  yamt     4464:        UVMMAP_EVCNT_INCR(uke_alloc);
1.174     yamt     4465:        entry = NULL;
                   4466: again:
                   4467:        /*
                   4468:         * try to grab an entry from freelist.
                   4469:         */
1.204.2.4  yamt     4470:        mutex_spin_enter(&uvm_kentry_lock);
1.175     yamt     4471:        ukh = LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free);
1.174     yamt     4472:        if (ukh) {
                   4473:                entry = uvm_kmapent_get(ukh);
                   4474:                if (ukh->ukh_nused == UVM_KMAPENT_CHUNK)
                   4475:                        LIST_REMOVE(ukh, ukh_listq);
                   4476:        }
1.204.2.4  yamt     4477:        mutex_spin_exit(&uvm_kentry_lock);
1.174     yamt     4478:
                   4479:        if (entry)
                   4480:                return entry;
                   4481:
                   4482:        /*
                   4483:         * there's no free entry for this vm_map.
                   4484:         * now we need to allocate some vm_map_entry.
                   4485:         * for simplicity, always allocate one page chunk of them at once.
                   4486:         */
                   4487:
                   4488:        pg = uvm_pagealloc(NULL, 0, NULL, 0);
                   4489:        if (__predict_false(pg == NULL)) {
                   4490:                if (flags & UVM_FLAG_NOWAIT)
                   4491:                        return NULL;
                   4492:                uvm_wait("kme_alloc");
                   4493:                goto again;
                   4494:        }
                   4495:
1.204.2.5  yamt     4496:        error = uvm_map_prepare(map, 0, PAGE_SIZE, NULL, UVM_UNKNOWN_OFFSET,
                   4497:            0, mapflags, &args);
1.174     yamt     4498:        if (error) {
                   4499:                uvm_pagefree(pg);
                   4500:                return NULL;
                   4501:        }
                   4502:
                   4503:        va = args.uma_start;
                   4504:
                   4505:        pmap_kenter_pa(va, VM_PAGE_TO_PHYS(pg), VM_PROT_READ|VM_PROT_WRITE);
                   4506:        pmap_update(vm_map_pmap(map));
                   4507:
                   4508:        ukh = (void *)va;
                   4509:
                   4510:        /*
                   4511:         * use the first entry for ukh itsself.
                   4512:         */
                   4513:
                   4514:        entry = &ukh->ukh_entries[0];
                   4515:        entry->flags = UVM_MAP_KERNEL | UVM_MAP_KMAPENT;
                   4516:        error = uvm_map_enter(map, &args, entry);
                   4517:        KASSERT(error == 0);
                   4518:
                   4519:        ukh->ukh_nused = UVM_KMAPENT_CHUNK;
                   4520:        ukh->ukh_map = map;
                   4521:        ukh->ukh_freelist = NULL;
                   4522:        for (i = UVM_KMAPENT_CHUNK - 1; i >= 2; i--) {
1.199     christos 4523:                struct vm_map_entry *xentry = &ukh->ukh_entries[i];
1.174     yamt     4524:
1.199     christos 4525:                xentry->flags = UVM_MAP_KERNEL;
                   4526:                uvm_kmapent_put(ukh, xentry);
1.174     yamt     4527:        }
                   4528:        KASSERT(ukh->ukh_nused == 2);
                   4529:
1.204.2.4  yamt     4530:        mutex_spin_enter(&uvm_kentry_lock);
1.175     yamt     4531:        LIST_INSERT_HEAD(&vm_map_to_kernel(map)->vmk_kentry_free,
                   4532:            ukh, ukh_listq);
1.204.2.4  yamt     4533:        mutex_spin_exit(&uvm_kentry_lock);
1.174     yamt     4534:
                   4535:        /*
                   4536:         * return second entry.
                   4537:         */
                   4538:
                   4539:        entry = &ukh->ukh_entries[1];
                   4540:        entry->flags = UVM_MAP_KERNEL;
1.204.2.1  yamt     4541:        UVMMAP_EVCNT_INCR(ukh_alloc);
1.174     yamt     4542:        return entry;
                   4543: }
                   4544:
                   4545: /*
                   4546:  * uvm_mapent_free: free map entry for in-kernel map
                   4547:  */
                   4548:
                   4549: static void
                   4550: uvm_kmapent_free(struct vm_map_entry *entry)
                   4551: {
                   4552:        struct uvm_kmapent_hdr *ukh;
                   4553:        struct vm_page *pg;
                   4554:        struct vm_map *map;
                   4555:        struct pmap *pmap;
                   4556:        vaddr_t va;
                   4557:        paddr_t pa;
                   4558:        struct vm_map_entry *deadentry;
                   4559:
1.204.2.1  yamt     4560:        UVMMAP_EVCNT_INCR(uke_free);
1.174     yamt     4561:        ukh = UVM_KHDR_FIND(entry);
                   4562:        map = ukh->ukh_map;
                   4563:
1.204.2.4  yamt     4564:        mutex_spin_enter(&uvm_kentry_lock);
1.174     yamt     4565:        uvm_kmapent_put(ukh, entry);
                   4566:        if (ukh->ukh_nused > 1) {
                   4567:                if (ukh->ukh_nused == UVM_KMAPENT_CHUNK - 1)
1.175     yamt     4568:                        LIST_INSERT_HEAD(
                   4569:                            &vm_map_to_kernel(map)->vmk_kentry_free,
                   4570:                            ukh, ukh_listq);
1.204.2.4  yamt     4571:                mutex_spin_exit(&uvm_kentry_lock);
1.174     yamt     4572:                return;
                   4573:        }
                   4574:
                   4575:        /*
                   4576:         * now we can free this ukh.
                   4577:         *
                   4578:         * however, keep an empty ukh to avoid ping-pong.
                   4579:         */
                   4580:
1.175     yamt     4581:        if (LIST_FIRST(&vm_map_to_kernel(map)->vmk_kentry_free) == ukh &&
1.174     yamt     4582:            LIST_NEXT(ukh, ukh_listq) == NULL) {
1.204.2.4  yamt     4583:                mutex_spin_exit(&uvm_kentry_lock);
1.174     yamt     4584:                return;
                   4585:        }
                   4586:        LIST_REMOVE(ukh, ukh_listq);
1.204.2.4  yamt     4587:        mutex_spin_exit(&uvm_kentry_lock);
1.174     yamt     4588:
                   4589:        KASSERT(ukh->ukh_nused == 1);
                   4590:
                   4591:        /*
                   4592:         * remove map entry for ukh itsself.
                   4593:         */
                   4594:
                   4595:        va = (vaddr_t)ukh;
                   4596:        KASSERT((va & PAGE_MASK) == 0);
1.198     yamt     4597:        vm_map_lock(map);
1.187     yamt     4598:        uvm_unmap_remove(map, va, va + PAGE_SIZE, &deadentry, NULL, 0);
1.174     yamt     4599:        KASSERT(deadentry->flags & UVM_MAP_KERNEL);
                   4600:        KASSERT(deadentry->flags & UVM_MAP_KMAPENT);
                   4601:        KASSERT(deadentry->next == NULL);
                   4602:        KASSERT(deadentry == &ukh->ukh_entries[0]);
                   4603:
                   4604:        /*
                   4605:         * unmap the page from pmap and free it.
                   4606:         */
                   4607:
                   4608:        pmap = vm_map_pmap(map);
                   4609:        KASSERT(pmap == pmap_kernel());
                   4610:        if (!pmap_extract(pmap, va, &pa))
                   4611:                panic("%s: no mapping", __func__);
                   4612:        pmap_kremove(va, PAGE_SIZE);
1.204.2.8  yamt     4613:        pmap_update(vm_map_pmap(map));
1.198     yamt     4614:        vm_map_unlock(map);
1.174     yamt     4615:        pg = PHYS_TO_VM_PAGE(pa);
                   4616:        uvm_pagefree(pg);
1.204.2.1  yamt     4617:        UVMMAP_EVCNT_INCR(ukh_free);
                   4618: }
                   4619:
                   4620: static vsize_t
                   4621: uvm_kmapent_overhead(vsize_t size)
                   4622: {
                   4623:
                   4624:        /*
                   4625:         * - the max number of unmerged entries is howmany(size, PAGE_SIZE)
                   4626:         *   as the min allocation unit is PAGE_SIZE.
                   4627:         * - UVM_KMAPENT_CHUNK "kmapent"s are allocated from a page.
                   4628:         *   one of them are used to map the page itself.
                   4629:         */
                   4630:
                   4631:        return howmany(howmany(size, PAGE_SIZE), (UVM_KMAPENT_CHUNK - 1)) *
                   4632:            PAGE_SIZE;
1.174     yamt     4633: }
                   4634:
                   4635: /*
                   4636:  * map entry reservation
                   4637:  */
                   4638:
                   4639: /*
                   4640:  * uvm_mapent_reserve: reserve map entries for clipping before locking map.
                   4641:  *
                   4642:  * => needed when unmapping entries allocated without UVM_FLAG_QUANTUM.
                   4643:  * => caller shouldn't hold map locked.
                   4644:  */
                   4645: int
                   4646: uvm_mapent_reserve(struct vm_map *map, struct uvm_mapent_reservation *umr,
                   4647:     int nentries, int flags)
                   4648: {
                   4649:
                   4650:        umr->umr_nentries = 0;
                   4651:
1.179     yamt     4652:        if ((flags & UVM_FLAG_QUANTUM) != 0)
                   4653:                return 0;
                   4654:
1.174     yamt     4655:        if (!VM_MAP_USE_KMAPENT(map))
                   4656:                return 0;
                   4657:
                   4658:        while (nentries--) {
                   4659:                struct vm_map_entry *ent;
                   4660:                ent = uvm_kmapent_alloc(map, flags);
                   4661:                if (!ent) {
                   4662:                        uvm_mapent_unreserve(map, umr);
                   4663:                        return ENOMEM;
                   4664:                }
                   4665:                UMR_PUTENTRY(umr, ent);
                   4666:        }
                   4667:
                   4668:        return 0;
                   4669: }
                   4670:
                   4671: /*
                   4672:  * uvm_mapent_unreserve:
                   4673:  *
                   4674:  * => caller shouldn't hold map locked.
                   4675:  * => never fail or sleep.
                   4676:  */
                   4677: void
                   4678: uvm_mapent_unreserve(struct vm_map *map, struct uvm_mapent_reservation *umr)
                   4679: {
                   4680:
                   4681:        while (!UMR_EMPTY(umr))
                   4682:                uvm_kmapent_free(UMR_GETENTRY(umr));
                   4683: }
                   4684:
1.194     yamt     4685: /*
                   4686:  * uvm_mapent_trymerge: try to merge an entry with its neighbors.
                   4687:  *
                   4688:  * => called with map locked.
                   4689:  * => return non zero if successfully merged.
                   4690:  */
                   4691:
                   4692: int
                   4693: uvm_mapent_trymerge(struct vm_map *map, struct vm_map_entry *entry, int flags)
                   4694: {
                   4695:        struct uvm_object *uobj;
                   4696:        struct vm_map_entry *next;
                   4697:        struct vm_map_entry *prev;
1.195     yamt     4698:        vsize_t size;
1.194     yamt     4699:        int merged = 0;
1.204.2.3  yamt     4700:        bool copying;
1.194     yamt     4701:        int newetype;
                   4702:
                   4703:        if (VM_MAP_USE_KMAPENT(map)) {
                   4704:                return 0;
                   4705:        }
                   4706:        if (entry->aref.ar_amap != NULL) {
                   4707:                return 0;
                   4708:        }
                   4709:        if ((entry->flags & UVM_MAP_NOMERGE) != 0) {
                   4710:                return 0;
                   4711:        }
                   4712:
                   4713:        uobj = entry->object.uvm_obj;
1.195     yamt     4714:        size = entry->end - entry->start;
1.194     yamt     4715:        copying = (flags & UVM_MERGE_COPYING) != 0;
                   4716:        newetype = copying ? (entry->etype & ~UVM_ET_NEEDSCOPY) : entry->etype;
                   4717:
                   4718:        next = entry->next;
                   4719:        if (next != &map->header &&
                   4720:            next->start == entry->end &&
                   4721:            ((copying && next->aref.ar_amap != NULL &&
                   4722:            amap_refs(next->aref.ar_amap) == 1) ||
                   4723:            (!copying && next->aref.ar_amap == NULL)) &&
                   4724:            UVM_ET_ISCOMPATIBLE(next, newetype,
                   4725:            uobj, entry->flags, entry->protection,
                   4726:            entry->max_protection, entry->inheritance, entry->advice,
1.195     yamt     4727:            entry->wired_count) &&
                   4728:            (uobj == NULL || entry->offset + size == next->offset)) {
1.194     yamt     4729:                int error;
                   4730:
                   4731:                if (copying) {
1.195     yamt     4732:                        error = amap_extend(next, size,
1.194     yamt     4733:                            AMAP_EXTEND_NOWAIT|AMAP_EXTEND_BACKWARDS);
                   4734:                } else {
                   4735:                        error = 0;
                   4736:                }
                   4737:                if (error == 0) {
1.197     yamt     4738:                        if (uobj) {
                   4739:                                if (uobj->pgops->pgo_detach) {
                   4740:                                        uobj->pgops->pgo_detach(uobj);
                   4741:                                }
1.194     yamt     4742:                        }
                   4743:
                   4744:                        entry->end = next->end;
1.204.2.1  yamt     4745:                        clear_hints(map, next);
1.194     yamt     4746:                        uvm_map_entry_unlink(map, next);
                   4747:                        if (copying) {
                   4748:                                entry->aref = next->aref;
                   4749:                                entry->etype &= ~UVM_ET_NEEDSCOPY;
                   4750:                        }
1.204.2.1  yamt     4751:                        uvm_map_check(map, "trymerge forwardmerge");
1.194     yamt     4752:                        uvm_mapent_free_merged(map, next);
                   4753:                        merged++;
                   4754:                }
                   4755:        }
                   4756:
                   4757:        prev = entry->prev;
                   4758:        if (prev != &map->header &&
                   4759:            prev->end == entry->start &&
                   4760:            ((copying && !merged && prev->aref.ar_amap != NULL &&
                   4761:            amap_refs(prev->aref.ar_amap) == 1) ||
                   4762:            (!copying && prev->aref.ar_amap == NULL)) &&
                   4763:            UVM_ET_ISCOMPATIBLE(prev, newetype,
                   4764:            uobj, entry->flags, entry->protection,
                   4765:            entry->max_protection, entry->inheritance, entry->advice,
1.195     yamt     4766:            entry->wired_count) &&
1.196     yamt     4767:            (uobj == NULL ||
                   4768:            prev->offset + prev->end - prev->start == entry->offset)) {
1.194     yamt     4769:                int error;
                   4770:
                   4771:                if (copying) {
1.195     yamt     4772:                        error = amap_extend(prev, size,
1.194     yamt     4773:                            AMAP_EXTEND_NOWAIT|AMAP_EXTEND_FORWARDS);
                   4774:                } else {
                   4775:                        error = 0;
                   4776:                }
                   4777:                if (error == 0) {
1.197     yamt     4778:                        if (uobj) {
                   4779:                                if (uobj->pgops->pgo_detach) {
                   4780:                                        uobj->pgops->pgo_detach(uobj);
                   4781:                                }
                   4782:                                entry->offset = prev->offset;
1.194     yamt     4783:                        }
                   4784:
                   4785:                        entry->start = prev->start;
1.204.2.1  yamt     4786:                        clear_hints(map, prev);
1.194     yamt     4787:                        uvm_map_entry_unlink(map, prev);
                   4788:                        if (copying) {
                   4789:                                entry->aref = prev->aref;
                   4790:                                entry->etype &= ~UVM_ET_NEEDSCOPY;
                   4791:                        }
1.204.2.1  yamt     4792:                        uvm_map_check(map, "trymerge backmerge");
1.194     yamt     4793:                        uvm_mapent_free_merged(map, prev);
                   4794:                        merged++;
                   4795:                }
                   4796:        }
                   4797:
                   4798:        return merged;
                   4799: }
                   4800:
1.1       mrg      4801: #if defined(DDB)
                   4802:
                   4803: /*
                   4804:  * DDB hooks
                   4805:  */
                   4806:
                   4807: /*
                   4808:  * uvm_map_printit: actually prints the map
                   4809:  */
                   4810:
1.10      mrg      4811: void
1.204.2.3  yamt     4812: uvm_map_printit(struct vm_map *map, bool full,
1.138     enami    4813:     void (*pr)(const char *, ...))
1.10      mrg      4814: {
1.99      chs      4815:        struct vm_map_entry *entry;
1.10      mrg      4816:
1.184     chs      4817:        (*pr)("MAP %p: [0x%lx->0x%lx]\n", map, vm_map_min(map),
                   4818:            vm_map_max(map));
1.53      thorpej  4819:        (*pr)("\t#ent=%d, sz=%d, ref=%d, version=%d, flags=0x%x\n",
                   4820:            map->nentries, map->size, map->ref_count, map->timestamp,
                   4821:            map->flags);
1.173     yamt     4822:        (*pr)("\tpmap=%p(resident=%ld, wired=%ld)\n", map->pmap,
                   4823:            pmap_resident_count(map->pmap), pmap_wired_count(map->pmap));
1.10      mrg      4824:        if (!full)
                   4825:                return;
                   4826:        for (entry = map->header.next; entry != &map->header;
                   4827:            entry = entry->next) {
1.70      kleink   4828:                (*pr)(" - %p: 0x%lx->0x%lx: obj=%p/0x%llx, amap=%p/%d\n",
1.10      mrg      4829:                    entry, entry->start, entry->end, entry->object.uvm_obj,
1.85      chs      4830:                    (long long)entry->offset, entry->aref.ar_amap,
                   4831:                    entry->aref.ar_pageoff);
1.10      mrg      4832:                (*pr)(
1.85      chs      4833:                    "\tsubmap=%c, cow=%c, nc=%c, prot(max)=%d/%d, inh=%d, "
                   4834:                    "wc=%d, adv=%d\n",
1.10      mrg      4835:                    (entry->etype & UVM_ET_SUBMAP) ? 'T' : 'F',
1.98      chs      4836:                    (entry->etype & UVM_ET_COPYONWRITE) ? 'T' : 'F',
1.10      mrg      4837:                    (entry->etype & UVM_ET_NEEDSCOPY) ? 'T' : 'F',
                   4838:                    entry->protection, entry->max_protection,
                   4839:                    entry->inheritance, entry->wired_count, entry->advice);
                   4840:        }
1.98      chs      4841: }
1.1       mrg      4842:
                   4843: /*
                   4844:  * uvm_object_printit: actually prints the object
                   4845:  */
                   4846:
1.10      mrg      4847: void
1.204.2.3  yamt     4848: uvm_object_printit(struct uvm_object *uobj, bool full,
1.138     enami    4849:     void (*pr)(const char *, ...))
1.10      mrg      4850: {
                   4851:        struct vm_page *pg;
                   4852:        int cnt = 0;
                   4853:
1.71      chs      4854:        (*pr)("OBJECT %p: locked=%d, pgops=%p, npages=%d, ",
1.204.2.7  yamt     4855:            uobj, mutex_owned(&uobj->vmobjlock), uobj->pgops, uobj->uo_npages);
1.42      thorpej  4856:        if (UVM_OBJ_IS_KERN_OBJECT(uobj))
1.10      mrg      4857:                (*pr)("refs=<SYSTEM>\n");
                   4858:        else
                   4859:                (*pr)("refs=%d\n", uobj->uo_refs);
                   4860:
1.71      chs      4861:        if (!full) {
                   4862:                return;
                   4863:        }
1.10      mrg      4864:        (*pr)("  PAGES <pg,offset>:\n  ");
1.106     chs      4865:        TAILQ_FOREACH(pg, &uobj->memq, listq) {
                   4866:                cnt++;
1.85      chs      4867:                (*pr)("<%p,0x%llx> ", pg, (long long)pg->offset);
1.106     chs      4868:                if ((cnt % 3) == 0) {
1.71      chs      4869:                        (*pr)("\n  ");
                   4870:                }
                   4871:        }
1.106     chs      4872:        if ((cnt % 3) != 0) {
1.71      chs      4873:                (*pr)("\n");
1.10      mrg      4874:        }
1.98      chs      4875: }
1.1       mrg      4876:
                   4877: /*
                   4878:  * uvm_page_printit: actually print the page
                   4879:  */
                   4880:
1.204.2.2  yamt     4881: static const char page_flagbits[] = UVM_PGFLAGBITS;
                   4882: static const char page_pqflagbits[] = UVM_PQFLAGBITS;
1.86      chs      4883:
1.10      mrg      4884: void
1.204.2.3  yamt     4885: uvm_page_printit(struct vm_page *pg, bool full,
1.138     enami    4886:     void (*pr)(const char *, ...))
1.10      mrg      4887: {
1.85      chs      4888:        struct vm_page *tpg;
1.10      mrg      4889:        struct uvm_object *uobj;
                   4890:        struct pglist *pgl;
1.71      chs      4891:        char pgbuf[128];
                   4892:        char pqbuf[128];
1.1       mrg      4893:
1.10      mrg      4894:        (*pr)("PAGE %p:\n", pg);
1.71      chs      4895:        bitmask_snprintf(pg->flags, page_flagbits, pgbuf, sizeof(pgbuf));
                   4896:        bitmask_snprintf(pg->pqflags, page_pqflagbits, pqbuf, sizeof(pqbuf));
1.106     chs      4897:        (*pr)("  flags=%s, pqflags=%s, wire_count=%d, pa=0x%lx\n",
1.137     yamt     4898:            pgbuf, pqbuf, pg->wire_count, (long)VM_PAGE_TO_PHYS(pg));
1.86      chs      4899:        (*pr)("  uobject=%p, uanon=%p, offset=0x%llx loan_count=%d\n",
                   4900:            pg->uobject, pg->uanon, (long long)pg->offset, pg->loan_count);
1.1       mrg      4901: #if defined(UVM_PAGE_TRKOWN)
1.10      mrg      4902:        if (pg->flags & PG_BUSY)
                   4903:                (*pr)("  owning process = %d, tag=%s\n",
                   4904:                    pg->owner, pg->owner_tag);
                   4905:        else
                   4906:                (*pr)("  page not busy, no owner\n");
1.1       mrg      4907: #else
1.10      mrg      4908:        (*pr)("  [page ownership tracking disabled]\n");
1.1       mrg      4909: #endif
                   4910:
1.10      mrg      4911:        if (!full)
                   4912:                return;
                   4913:
                   4914:        /* cross-verify object/anon */
                   4915:        if ((pg->pqflags & PQ_FREE) == 0) {
                   4916:                if (pg->pqflags & PQ_ANON) {
1.192     yamt     4917:                        if (pg->uanon == NULL || pg->uanon->an_page != pg)
1.85      chs      4918:                            (*pr)("  >>> ANON DOES NOT POINT HERE <<< (%p)\n",
1.192     yamt     4919:                                (pg->uanon) ? pg->uanon->an_page : NULL);
1.10      mrg      4920:                        else
                   4921:                                (*pr)("  anon backpointer is OK\n");
                   4922:                } else {
                   4923:                        uobj = pg->uobject;
                   4924:                        if (uobj) {
                   4925:                                (*pr)("  checking object list\n");
1.85      chs      4926:                                TAILQ_FOREACH(tpg, &uobj->memq, listq) {
                   4927:                                        if (tpg == pg) {
                   4928:                                                break;
                   4929:                                        }
1.10      mrg      4930:                                }
1.85      chs      4931:                                if (tpg)
1.10      mrg      4932:                                        (*pr)("  page found on object list\n");
                   4933:                                else
                   4934:                        (*pr)("  >>> PAGE NOT FOUND ON OBJECT LIST! <<<\n");
                   4935:                        }
                   4936:                }
                   4937:        }
1.1       mrg      4938:
1.10      mrg      4939:        /* cross-verify page queue */
1.73      thorpej  4940:        if (pg->pqflags & PQ_FREE) {
                   4941:                int fl = uvm_page_lookup_freelist(pg);
1.96      thorpej  4942:                int color = VM_PGCOLOR_BUCKET(pg);
                   4943:                pgl = &uvm.page_free[fl].pgfl_buckets[color].pgfl_queues[
                   4944:                    ((pg)->flags & PG_ZERO) ? PGFL_ZEROS : PGFL_UNKNOWN];
1.139     enami    4945:        } else {
1.10      mrg      4946:                pgl = NULL;
1.85      chs      4947:        }
1.10      mrg      4948:
                   4949:        if (pgl) {
                   4950:                (*pr)("  checking pageq list\n");
1.85      chs      4951:                TAILQ_FOREACH(tpg, pgl, pageq) {
                   4952:                        if (tpg == pg) {
                   4953:                                break;
                   4954:                        }
1.10      mrg      4955:                }
1.85      chs      4956:                if (tpg)
1.10      mrg      4957:                        (*pr)("  page found on pageq list\n");
                   4958:                else
                   4959:                        (*pr)("  >>> PAGE NOT FOUND ON PAGEQ LIST! <<<\n");
                   4960:        }
1.1       mrg      4961: }
1.204.2.1  yamt     4962:
                   4963: /*
                   4964:  * uvm_pages_printthem - print a summary of all managed pages
                   4965:  */
                   4966:
                   4967: void
                   4968: uvm_page_printall(void (*pr)(const char *, ...))
                   4969: {
                   4970:        unsigned i;
                   4971:        struct vm_page *pg;
                   4972:
1.204.2.2  yamt     4973:        (*pr)("%18s %4s %4s %18s %18s"
1.204.2.1  yamt     4974: #ifdef UVM_PAGE_TRKOWN
                   4975:            " OWNER"
                   4976: #endif
                   4977:            "\n", "PAGE", "FLAG", "PQ", "UOBJECT", "UANON");
                   4978:        for (i = 0; i < vm_nphysseg; i++) {
                   4979:                for (pg = vm_physmem[i].pgs; pg <= vm_physmem[i].lastpg; pg++) {
1.204.2.2  yamt     4980:                        (*pr)("%18p %04x %04x %18p %18p",
1.204.2.1  yamt     4981:                            pg, pg->flags, pg->pqflags, pg->uobject,
                   4982:                            pg->uanon);
                   4983: #ifdef UVM_PAGE_TRKOWN
                   4984:                        if (pg->flags & PG_BUSY)
                   4985:                                (*pr)(" %d [%s]", pg->owner, pg->owner_tag);
                   4986: #endif
                   4987:                        (*pr)("\n");
                   4988:                }
                   4989:        }
                   4990: }
                   4991:
1.1       mrg      4992: #endif
1.204.2.1  yamt     4993:
                   4994: /*
                   4995:  * uvm_map_create: create map
                   4996:  */
                   4997:
                   4998: struct vm_map *
                   4999: uvm_map_create(pmap_t pmap, vaddr_t vmin, vaddr_t vmax, int flags)
                   5000: {
                   5001:        struct vm_map *result;
                   5002:
                   5003:        MALLOC(result, struct vm_map *, sizeof(struct vm_map),
                   5004:            M_VMMAP, M_WAITOK);
                   5005:        uvm_map_setup(result, vmin, vmax, flags);
                   5006:        result->pmap = pmap;
                   5007:        return(result);
                   5008: }
                   5009:
                   5010: /*
                   5011:  * uvm_map_setup: init map
                   5012:  *
                   5013:  * => map must not be in service yet.
                   5014:  */
                   5015:
                   5016: void
                   5017: uvm_map_setup(struct vm_map *map, vaddr_t vmin, vaddr_t vmax, int flags)
                   5018: {
1.204.2.4  yamt     5019:        int ipl;
1.204.2.1  yamt     5020:
                   5021:        RB_INIT(&map->rbhead);
                   5022:        map->header.next = map->header.prev = &map->header;
                   5023:        map->nentries = 0;
                   5024:        map->size = 0;
                   5025:        map->ref_count = 1;
                   5026:        vm_map_setmin(map, vmin);
                   5027:        vm_map_setmax(map, vmax);
                   5028:        map->flags = flags;
                   5029:        map->first_free = &map->header;
                   5030:        map->hint = &map->header;
                   5031:        map->timestamp = 0;
1.204.2.4  yamt     5032:        map->busy = NULL;
                   5033:
                   5034:        if ((flags & VM_MAP_INTRSAFE) != 0) {
                   5035:                ipl = IPL_VM;
                   5036:        } else {
                   5037:                ipl = IPL_NONE;
                   5038:        }
                   5039:
                   5040:        rw_init(&map->lock);
                   5041:        cv_init(&map->cv, "vm_map");
                   5042:        mutex_init(&map->misc_lock, MUTEX_DRIVER, ipl);
                   5043:        mutex_init(&map->mutex, MUTEX_DRIVER, ipl);
1.204.2.1  yamt     5044: }
                   5045:
                   5046:
                   5047: /*
                   5048:  *   U N M A P   -   m a i n   e n t r y   p o i n t
                   5049:  */
                   5050:
                   5051: /*
                   5052:  * uvm_unmap1: remove mappings from a vm_map (from "start" up to "stop")
                   5053:  *
                   5054:  * => caller must check alignment and size
                   5055:  * => map must be unlocked (we will lock it)
                   5056:  * => flags is UVM_FLAG_QUANTUM or 0.
                   5057:  */
                   5058:
                   5059: void
                   5060: uvm_unmap1(struct vm_map *map, vaddr_t start, vaddr_t end, int flags)
                   5061: {
                   5062:        struct vm_map_entry *dead_entries;
                   5063:        struct uvm_mapent_reservation umr;
                   5064:        UVMHIST_FUNC("uvm_unmap"); UVMHIST_CALLED(maphist);
                   5065:
                   5066:        UVMHIST_LOG(maphist, "  (map=0x%x, start=0x%x, end=0x%x)",
                   5067:            map, start, end, 0);
1.204.2.6  yamt     5068:        if (map == kernel_map) {
                   5069:                LOCKDEBUG_MEM_CHECK((void *)start, end - start);
                   5070:        }
1.204.2.1  yamt     5071:        /*
                   5072:         * work now done by helper functions.   wipe the pmap's and then
                   5073:         * detach from the dead entries...
                   5074:         */
                   5075:        uvm_mapent_reserve(map, &umr, 2, flags);
                   5076:        vm_map_lock(map);
                   5077:        uvm_unmap_remove(map, start, end, &dead_entries, &umr, flags);
                   5078:        vm_map_unlock(map);
                   5079:        uvm_mapent_unreserve(map, &umr);
                   5080:
                   5081:        if (dead_entries != NULL)
                   5082:                uvm_unmap_detach(dead_entries, 0);
                   5083:
                   5084:        UVMHIST_LOG(maphist, "<- done", 0,0,0,0);
                   5085: }
                   5086:
                   5087:
                   5088: /*
                   5089:  * uvm_map_reference: add reference to a map
                   5090:  *
1.204.2.4  yamt     5091:  * => map need not be locked (we use misc_lock).
1.204.2.1  yamt     5092:  */
                   5093:
                   5094: void
                   5095: uvm_map_reference(struct vm_map *map)
                   5096: {
1.204.2.4  yamt     5097:        mutex_enter(&map->misc_lock);
1.204.2.1  yamt     5098:        map->ref_count++;
1.204.2.4  yamt     5099:        mutex_exit(&map->misc_lock);
1.204.2.1  yamt     5100: }
                   5101:
                   5102: struct vm_map_kernel *
                   5103: vm_map_to_kernel(struct vm_map *map)
                   5104: {
                   5105:
                   5106:        KASSERT(VM_MAP_IS_KERNEL(map));
                   5107:
                   5108:        return (struct vm_map_kernel *)map;
                   5109: }
                   5110:
1.204.2.3  yamt     5111: bool
1.204.2.1  yamt     5112: vm_map_starved_p(struct vm_map *map)
                   5113: {
                   5114:
                   5115:        if ((map->flags & VM_MAP_WANTVA) != 0) {
1.204.2.3  yamt     5116:                return true;
1.204.2.1  yamt     5117:        }
                   5118:        /* XXX */
                   5119:        if ((vm_map_max(map) - vm_map_min(map)) / 16 * 15 < map->size) {
1.204.2.3  yamt     5120:                return true;
1.204.2.1  yamt     5121:        }
1.204.2.3  yamt     5122:        return false;
1.204.2.1  yamt     5123: }
1.204.2.7  yamt     5124:
                   5125: #if defined(DDB)
                   5126: void
                   5127: uvm_whatis(uintptr_t addr, void (*pr)(const char *, ...))
                   5128: {
                   5129:        struct vm_map *map;
                   5130:
                   5131:        for (map = kernel_map;;) {
                   5132:                struct vm_map_entry *entry;
                   5133:
                   5134:                if (!uvm_map_lookup_entry_bytree(map, (vaddr_t)addr, &entry)) {
                   5135:                        break;
                   5136:                }
                   5137:                (*pr)("%p is %p+%zu from VMMAP %p\n",
                   5138:                    (void *)addr, (void *)entry->start,
                   5139:                    (size_t)(addr - (uintptr_t)entry->start), map);
                   5140:                if (!UVM_ET_ISSUBMAP(entry)) {
                   5141:                        break;
                   5142:                }
                   5143:                map = entry->object.sub_map;
                   5144:        }
                   5145: }
                   5146: #endif /* defined(DDB) */

CVSweb <webmaster@jp.NetBSD.org>